LSL-PyOptimizer/lslopt/lslparse.py

3085 lines
125 KiB
Python
Raw Permalink Normal View History

2024-04-14 02:40:21 -07:00
# (C) Copyright 2015-2024 Sei Lisa. All rights reserved.
#
# This file is part of LSL PyOptimizer.
#
# LSL PyOptimizer is free software: you can redistribute it and/or
# modify it under the terms of the GNU General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# LSL PyOptimizer is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with LSL PyOptimizer. If not, see <http://www.gnu.org/licenses/>.
# Parser module. Converts the source into an abstract syntax tree,
# generating also the symbol table.
2014-08-06 13:49:57 -07:00
# TODO: Add info to be able to propagate error position to the source.
from lslopt.lslcommon import Key, Vector, Quaternion, types, nr
from lslopt import lslcommon, lslfuncs
from strutil import *
strutil_used
import re
# Note this module was basically written from bottom to top, which may help
# reading it.
2020-11-08 18:28:57 -07:00
WHITESPACE_CHARS = frozenset({' ', '\r', '\n', '\x0B', '\x0C'})
SINGLE_SYMBOLS = frozenset({'.', ';', '{', '}', ',', '=', '(', ')', '-', '+',
'*', '/', '%', '@', ':', '<', '>', '[', ']', '&', '|', '^', '~', '!'})
def isdigit(c):
return '0' <= c <= '9'
def isalpha_(c):
return c == '_' or 'A' <= c <= 'Z' or 'a' <= c <= 'z'
def isalphanum_(c):
return isalpha_(c) or isdigit(c)
def ishex(c):
return '0' <= c <= '9' or 'A' <= c <= 'F' or 'a' <= c <= 'f'
def GetErrLineCol(parser):
errorpos = parser.errorpos
# Find zero-based line number
lno = parser.script.count('\n', 0, errorpos)
# Find start of current line
lstart = parser.script.rfind('\n', 0, errorpos) + 1
# Find zero-based column number in characters
2020-11-08 18:28:57 -07:00
cno = len(any2u(parser.script[lstart:errorpos], 'utf8'))
# Find in #line directives list
i = len(parser.linedir)
filename = '<stdin>' # value to return if there's no #line before lno
while i:
i -= 1
line = parser.linedir[i]
# We wouldn't know where to report the error in this case:
assert lno != line[0], \
"Error position is in processed #line directive?!"
if line[0] < lno: # found the last #line directive before lno
# replace the value of lno
lno = lno - line[0] + line[1] - 2
filename = line[2]
break
return (lno + 1, cno + 1, filename)
class EParse(Exception):
def __init__(self, parser, msg):
self.errorpos = parser.errorpos
self.lno, self.cno, self.fname = GetErrLineCol(parser)
filename = self.fname
if parser.emap and filename == '<stdin>':
filename = parser.filename
2020-11-08 18:28:57 -07:00
filename = (str2u(filename, 'utf8')
.replace(u'\\', u'\\\\')
.replace(u'"', u'\\"')
)
if parser.emap:
msg = u'::ERROR::"%s":%d:%d: %s' % (
any2u(filename.lstrip('u')), self.lno, self.cno, msg)
elif parser.processpre and filename != '<stdin>':
msg = u"(Line %d char %d): ERROR in \"%s\": %s" % (
self.lno, self.cno, filename, msg)
else:
msg = u"(Line %d char %d): ERROR: %s" % (self.lno, self.cno, msg)
super(EParse, self).__init__(msg)
class EParseUEOF(EParse):
def __init__(self, parser):
parser.errorpos = len(parser.script)
super(EParseUEOF, self).__init__(parser, u"Unexpected EOF")
class EParseSyntax(EParse):
def __init__(self, parser):
super(EParseSyntax, self).__init__(parser, u"Syntax error")
class EParseAlreadyDefined(EParse):
def __init__(self, parser):
super(EParseAlreadyDefined, self).__init__(parser,
u"Name previously declared within scope")
class EParseUndefined(EParse):
def __init__(self, parser):
super(EParseUndefined, self).__init__(parser,
u"Name not defined within scope")
class EParseTypeMismatch(EParse):
def __init__(self, parser):
super(EParseTypeMismatch, self).__init__(parser, u"Type mismatch")
class EParseReturnShouldBeEmpty(EParse):
def __init__(self, parser):
# When the types don't match, the error es EParseTypeMismatch instead.
super(EParseReturnShouldBeEmpty, self).__init__(parser,
u"Return statement type doesn't match function return type")
class EParseReturnIsEmpty(EParse):
def __init__(self, parser):
2014-07-30 16:34:34 -07:00
super(EParseReturnIsEmpty, self).__init__(parser,
u"Function returns a value but return statement doesn't")
# This error message may sound funny, for good reasons.
class EParseInvalidField(EParse):
def __init__(self, parser):
super(EParseInvalidField, self).__init__(parser,
u"Use of vector or quaternion method on incorrect type")
class EParseFunctionMismatch(EParse):
def __init__(self, parser):
super(EParseFunctionMismatch, self).__init__(parser,
2015-03-12 19:28:51 -07:00
u"Function call mismatches type or number of arguments")
class EParseDeclarationScope(EParse):
def __init__(self, parser):
super(EParseDeclarationScope, self).__init__(parser,
u"Declaration requires a new scope -- use { and }")
class EParseCantChangeState(EParse):
def __init__(self, parser):
super(EParseCantChangeState, self).__init__(parser,
u"Global functions can't change state")
class EParseCodePathWithoutRet(EParse):
def __init__(self, parser):
super(EParseCodePathWithoutRet, self).__init__(parser,
u"Not all code paths return a value")
class EParseDuplicateLabel(EParse):
def __init__(self, parser):
super(EParseDuplicateLabel, self).__init__(parser,
u"Duplicate local label name. That won't allow the Mono script"
u" to be saved, and will not work as expected in LSO.")
class EParseInvalidCase(EParse):
def __init__(self, parser, kind):
super(EParseInvalidCase, self).__init__(parser,
u"'%s' used outside a 'switch' statement" % kind)
class EParseCaseNotAllowed(EParse):
def __init__(self, parser, kind):
super(EParseCaseNotAllowed, self).__init__(parser,
u"'%s' label only allowed at the main 'switch' block" % kind)
class EParseManyDefaults(EParse):
def __init__(self, parser):
super(EParseManyDefaults, self).__init__(parser,
u"multiple 'default' labels inside 'switch' statement")
class EParseMissingDefault(EParse):
def __init__(self, parser):
super(EParseMissingDefault, self).__init__(parser,
u"Missing 'default:' label inside 'switch' statement; disable"
u" option 'errmissingdefault' to disable this error.")
class EParseInvalidBreak(EParse):
def __init__(self, parser):
super(EParseInvalidBreak, self).__init__(parser,
u"'break' used outside a loop or switch"
if parser.enableswitch and parser.breakcont
else u"'break' used outside a switch" if parser.enableswitch
else u"'break' used outside a loop")
class EParseInvalidCont(EParse):
def __init__(self, parser):
super(EParseInvalidCont, self).__init__(parser,
u"'continue' used outside a loop")
class EParseInvalidBrkContArg(EParse):
def __init__(self, parser):
super(EParseInvalidBrkContArg, self).__init__(parser,
u"Invalid argument to 'break' or 'continue'" if parser.breakcont
else u"Invalid argument to 'break'")
class EParseInvalidBackslash(EParse):
def __init__(self, parser):
super(EParseInvalidBackslash, self).__init__(parser,
2016-06-27 13:38:19 -07:00
u"Preprocessor directive can't end in backslash."
u" Activate the preprocessor or put everything in the same line.")
class EParseInvalidLabelOpt(EParse):
def __init__(self, parser):
super(EParseInvalidLabelOpt, self).__init__(parser,
u"When optimization is active, a label can't be the immediate"
u" child of a 'for', 'if', 'while' or 'do'. Disable optimization"
u" or rewrite the code in some other way.")
class EParseNoConversion(EParse):
def __init__(self, parser):
super(EParseNoConversion, self).__init__(parser,
u"There's no conversion function in the library for this type")
class EInternal(Exception):
"""This exception is a construct to allow a different function to cause an
2015-03-26 15:26:36 -07:00
immediate return of EOF from parser.GetToken().
"""
pass
class parser(object):
2020-11-08 18:06:37 -07:00
assignment_toks = frozenset({'=', '+=', '-=', '*=', '/=', '%='})
extassignment_toks = frozenset({'|=', '&=', '^=', '<<=', '>>='})
2020-11-08 18:06:37 -07:00
double_toks = frozenset({'++', '--', '+=', '-=', '*=', '/=', '%=', '==',
'!=', '>=', '<=', '&&', '||', '<<', '>>'})
extdouble_toks = frozenset({'|=', '&=', '^='})
# These are hardcoded because additions or modifications imply
# important changes to the code anyway.
2020-11-08 18:06:37 -07:00
base_keywords = frozenset({'default', 'state', 'event', 'jump', 'return',
'if', 'else', 'for', 'do', 'while', 'print', 'TRUE', 'FALSE'})
brkcont_keywords = frozenset({'break', 'continue'})
switch_keywords = frozenset({'switch', 'case', 'break', 'default'})
PythonType2LSLToken = {int:'INTEGER_VALUE', float:'FLOAT_VALUE',
unicode:'STRING_VALUE', Key:'KEY_VALUE', Vector:'VECTOR_VALUE',
Quaternion:'ROTATION_VALUE', list:'LIST_VALUE'}
TypeToExtractionFunction = {}
# Utility function
def GenerateLabel(self):
while True:
self.labelcnt += 1
unique = 'J_autoGen%05d' % self.labelcnt
if unique not in self.locallabels:
break
self.locallabels.add(unique)
return unique
def PushScope(self):
"""Create a new symbol table / scope level"""
self.scopeindex = len(self.symtab)
self.symtab.append({}) # Add new symbol table
self.scopestack.append(self.scopeindex)
def PopScope(self):
"""Return to the previous scope level"""
assert self.scopeindex == self.scopestack[-1]
self.scopestack.pop()
self.scopeindex = self.scopestack[-1]
assert len(self.scopestack) > 0
def AddSymbol(self, kind, scope, name, **values):
values['Kind'] = kind
if kind in ('v', 'l'):
values['Scope'] = scope
self.symtab[scope][name] = values
def FindSymbolPartial(self, symbol, MustBeLabel = False):
"""Find a symbol in all visible scopes in order, but not in the full
globals table (only globals seen so far are visible).
Labels have special scope rules: other identifiers with the same
name that are not labels are invisible to JUMP statements. Example:
default{timer(){ @x; {integer x; jump x;} }}
finds the label at the outer block. However:
default{timer(){ @x; integer x; }}
gives an identifier already defined error. On the other hand, labels
hide other types (but that's dealt with in the caller to this function):
default{timer(){ integer a; { @a; a++; } }}
gives an Name Not Defined error.
"""
scopelevel = len(self.scopestack)
while scopelevel:
scopelevel -= 1
symtab = self.symtab[self.scopestack[scopelevel]]
if symbol in symtab and (not MustBeLabel
or symtab[symbol]['Kind'] == 'l'):
return symtab[symbol]
return None
# No labels or states allowed here (but functions are)
def FindSymbolFull(self, symbol, globalonly=False):
"""Returns the symbol table entry for the given symbol."""
scopelevel = 1 if globalonly else len(self.scopestack)
while scopelevel: # Loop over all scopes in the stack
scopelevel -= 1
symtab = self.symtab[self.scopestack[scopelevel]]
if symbol in symtab:
# This can't happen, as functions can't be local
#if len(symtab[symbol]) > 3:
# return (symtab[symbol][1], symtab[symbol][3])
return symtab[symbol]
try:
return self.symtab[0][symbol] # Quick guess
except KeyError:
if (self.disallowglobalvars and symbol not in self.symtab[0]
or symbol not in self.globals
):
return None # Disallow forwards in global var mode
return self.globals[symbol]
def ValidateField(self, typ, field):
if typ == 'vector' and field in ('x', 'y', 'z') \
or typ == 'rotation' and field in ('x', 'y', 'z', 's'):
return
raise EParseInvalidField(self)
def autocastcheck(self, value, tgttype):
"""Check if automatic dynamic cast is possible. If explicit casts are
requested, insert one.
"""
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
tval = value.t
if tval == tgttype:
return value
if tval in ('string', 'key') and tgttype in ('string', 'key') \
or tval == 'integer' and tgttype == 'float':
if self.explicitcast:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='CAST', t=tgttype, ch=[value])
return value
raise EParseTypeMismatch(self)
def ueof(self):
"""Check for unexpected EOF"""
if self.pos >= self.length:
raise EParseUEOF(self)
def ceof(self):
"""Check for normal EOF"""
if self.pos >= self.length:
raise EInternal() # force GetToken to return EOF
def SetOpt(self, option, value):
# See parse() for meaning of options.
if option == 'extendedglobalexpr':
self.extendedglobalexpr = value
if option == 'extendedtypecast':
self.extendedtypecast = value
if option == 'extendedassignment':
self.extendedassignment = value
if option == 'explicitcast':
self.explicitcast = value
if option == 'allowkeyconcat':
self.allowkeyconcat = value
if option == 'allowmultistrings':
self.allowmultistrings = value
if option == 'processpre':
self.processpre = value
# TODO: Allow pure C-style string escapes. This is low-priority.
#if option == 'allowcescapes':
# self.allowcescapes = value
# Enable switch statements.
if option == 'enableswitch':
if not self.enableswitch and value:
self.keywords |= self.switch_keywords
elif self.enableswitch and not value:
self.keywords = self.base_keywords.copy()
if self.breakcont:
self.keywords |= self.brkcont_keywords
self.enableswitch = value
# Enable break/continue
if option == 'breakcont':
if not self.breakcont and value:
self.keywords |= self.brkcont_keywords
elif self.breakcont and not value:
self.keywords = self.base_keywords.copy()
if self.enableswitch:
self.keywords |= self.switch_keywords
self.breakcont = value
if option == 'errmissingdefault':
self.errmissingdefault = value
if option == 'lazylists':
self.lazylists = value
if option == 'duplabels':
self.duplabels = value
if option == 'shrinknames':
self.shrinknames = value
if option == 'funcoverride':
self.funcoverride = value
2019-01-05 18:20:34 -07:00
if option == 'inline':
self.enable_inline = value
def ProcessDirective(self, directive):
"""Process a given preprocessor directive during parsing."""
# Ignore directives on the first pass
if self.scanglobals:
return
if directive[len(directive)-1:] == '\\':
raise EParseInvalidBackslash(self)
# compile the RE lazily, to avoid penalizing programs not using it
if self.parse_directive_re is None:
self.parse_directive_re = re.compile(
r'^#\s*(?:'
r'(?:line)?\s+(\d+)(?:\s+("(?:\\.|[^"])*")(?:\s+\d+)*)?'
r'|'
r'(?:pragma)\s+(?:OPT)\s+([-+,a-z0-9_]+)'
r'|'
r'([a-z0-9_]+)(?:\s+(.*)?)' # others
r')\s*$'
, re.I
)
match = self.parse_directive_re.search(directive)
if match is not None:
# Something parsed
if match.group(1) is not None:
#line directive
if match.group(2) is not None:
# filename included
if match.group(2).find('\\') != -1:
# interpret escapes
from ast import literal_eval
filename = literal_eval(match.group(2))
else:
filename = match.group(2)[1:-1]
self.lastFILE = filename
else:
filename = self.lastFILE
# Referenced line number (in the #line directive)
reflinenum = int(match.group(1))
# Actual line number (where the #line directive itself is)
# FIXME: this is O(n^2); track line number instead of this hack
actlinenum = self.script.count('\n', 0, self.pos)
self.linedir.append((actlinenum, reflinenum, filename))
del actlinenum, reflinenum, filename
elif match.group(3): # '#pragma OPT <options>' found
opts = match.group(3).lower().split(',')
for opt in opts:
if opt != '':
if opt[0] == '-':
self.SetOpt(opt[1:], False)
elif opt[0] == '+':
self.SetOpt(opt[1:], True)
else:
self.SetOpt(opt, True)
elif match.group(4) == 'warning':
if match.group(5):
warning("Warning: #warning " + match.group(5))
else:
warning("Warning: #warning")
# else ignore
def GetToken(self):
"""Lexer"""
try:
while self.pos < self.length:
# In case of error, report it at the start of this token.
self.errorpos = self.pos
c = self.script[self.pos]
self.pos += 1
# Process preprocessor directives
if self.processpre and self.linestart and c == '#':
# Preprocessor directive.
# Most are not supposed to reach us but some do:
# - gcpp generates lines in the output like:
# # 123 "file.lsl"
# - other preprocessors including Boost Wave and mcpp
# generate lines like:
# #line 123 "file.lsl"
# Firestorm comments these out and instead outputs
# //#line 123 "file.lsl"
# - #pragma directives
# - #define directives from mcpp's #pragma MCPP put_defines
# or from gcpp's -dN option, that we use to detect some
# definitions.
self.ceof()
while self.script[self.pos] != '\n':
self.pos += 1
self.ceof() # A preprocessor command at EOF is not unexpected EOF.
self.ProcessDirective(self.script[self.errorpos:self.pos])
self.pos += 1
self.ceof()
continue
# Process comments
if c == '/':
if self.script[self.pos:self.pos+1] == '/':
self.pos += 1
if self.enable_inline and self.script.startswith(
'pragma inline', self.pos
) and not isalphanum_(self.script[self.pos + 13:
self.pos + 14]
):
self.pos += 12 # len('pragma inline') - 1
while self.script[self.pos] != '\n':
self.pos += 1
# Check for normal EOF. Note: 'inline' is not
# inserted if the file ends before a newline.
self.ceof()
return ('IDENT', 'inline')
self.ceof()
while self.script[self.pos] != '\n':
self.pos += 1
self.ceof() # A single-line comment at EOF is not unexpected EOF.
self.linestart = True
self.pos += 1
self.ceof()
continue
elif self.script[self.pos:self.pos+1] == '*':
self.pos += 2
if self.enable_inline and self.script.startswith(
'pragma inline*/', self.pos-1):
self.pos += 14 # len('pragma inline*/') - 1
return ('IDENT', 'inline')
while self.script[self.pos-1:self.pos+1] != '*/':
self.pos += 1
self.ueof() # An unterminated multiline comment *is* unexpected EOF.
self.pos += 1
self.ceof()
continue
# self.linestart is related to the preprocessor, therefore we
# check the characters that are relevant for standard C.
2020-11-08 18:28:57 -07:00
if c not in WHITESPACE_CHARS:
self.linestart = False
# Process strings
if c == '"' or c == 'L' and self.script[self.pos:self.pos+1] == '"':
strliteral = ''
if c == 'L':
self.pos += 1
strliteral = '"'
savepos = self.pos # we may need to backtrack
is_string = True # by default
while self.script[self.pos:self.pos+1] != '"':
# per the grammar, on EOF, it's not considered a string
if self.pos >= self.length:
self.pos = savepos
is_string = False
break
if self.script[self.pos] == '\\':
self.pos += 1
self.ueof()
if self.script[self.pos] == 'n':
strliteral += '\n'
elif self.script[self.pos] == 't':
strliteral += ' '
elif self.script[self.pos] == '\n':
# '\' followed by a newline; it's not a string.
self.pos = savepos
is_string = False
self.linestart = True
break
else:
strliteral += self.script[self.pos]
else:
strliteral += self.script[self.pos]
self.pos += 1
if is_string:
self.pos += 1
2020-11-08 18:28:57 -07:00
return ('STRING_VALUE', lslfuncs.zstr(str2u(strliteral, 'utf8')))
# fall through (to consider the L or to ignore the ")
if isalpha_(c):
# Identifier or reserved
ident = c
while isalphanum_(self.script[self.pos:self.pos+1]):
ident += self.script[self.pos]
self.pos += 1
# Got an identifier - check if it's a reserved word
if ident in self.keywords:
return (ident.upper(),)
if ident in types:
if ident == 'quaternion':
ident = 'rotation' # Normalize types
return ('TYPE',ident)
if ident in self.events:
return ('EVENT_NAME',ident)
if ident in self.constants:
value = self.constants[ident]
return (self.PythonType2LSLToken[type(value)], value)
return ('IDENT', ident)
# Process numbers: float, hex integer, dec integer
if c == '.' or isdigit(c):
number = ''
if c != '.':
# We have a digit, which means we have for sure either
# an integer or a float.
# Eat as many decimal digits as possible
number = c
while isdigit(self.script[self.pos:self.pos+1]):
number += self.script[self.pos]
self.pos += 1
if number == '0' and self.script[self.pos:self.pos+1] in ('x','X') \
and ishex(self.script[self.pos+1:self.pos+2]):
# We don't need the 0x prefix.
self.pos += 1
# Eat leading zeros to know the real length.
while self.script[self.pos:self.pos+1] == '0':
self.pos += 1
number = ''
while ishex(self.script[self.pos:self.pos+1]):
if len(number) < 9: # don't let it grow more than necessary
number += self.script[self.pos]
self.pos += 1
if number == '':
# We know there was at least a valid digit so it
# must've been all zeros.
number = '0'
if len(number) > 8:
number = -1
else:
number = lslfuncs.S32(int(number, 16))
return ('INTEGER_VALUE', number)
# Add the dot if present
if self.script[self.pos:self.pos+1] == '.':
number += '.'
self.pos += 1
else:
number = c
while isdigit(self.script[self.pos:self.pos+1]):
number += self.script[self.pos]
self.pos += 1
# At this point, number contains as many digits as there are before the dot,
# the dot if present, and as many digits as there are after the dot.
if number != '.': # A dot alone can't be a number so we rule it out here.
exp = ''
if self.script[self.pos:self.pos+1] in ('e','E'):
epos = self.pos # Temporary position tracker, made permanent only if the match succeeds
exp = self.script[epos]
epos += 1
if self.script[epos:epos+1] in ('+','-'):
exp += self.script[epos]
epos += 1
if isdigit(self.script[epos:epos+1]):
# Now we *do* have an exponent.
exp += self.script[epos]
epos += 1
while isdigit(self.script[epos:epos+1]):
exp += self.script[epos]
epos += 1
self.pos = epos # "Commit" the new position
else:
exp = '' # No cigar. Rollback and backtrack. Invalidate exp.
if exp != '' or '.' in number: # Float
if '.' in number:
# Eat the 'F' if present
if self.script[self.pos:self.pos+1] in ('f','F'):
# Python doesn't like the 'F' so don't return it
#exp += self.script[self.pos]
self.pos += 1
return ('FLOAT_VALUE', lslfuncs.F32(float(number + exp)))
if len(number) > 10 or len(number) == 10 and number > '4294967295':
number = -1
else:
number = lslfuncs.S32(int(number))
return ('INTEGER_VALUE', number)
if self.script[self.pos-1:self.pos+1] in self.double_toks \
or self.extendedassignment and self.script[self.pos-1:self.pos+1] in self.extdouble_toks:
self.pos += 1
if self.extendedassignment and self.script[self.pos-2:self.pos+1] in ('<<=', '>>='):
self.pos += 1
return (self.script[self.pos-3:self.pos],)
return (self.script[self.pos-2:self.pos],)
2020-11-08 18:28:57 -07:00
if c in SINGLE_SYMBOLS:
return (c,)
if c == '\n':
self.linestart = True
2015-03-09 05:20:37 -07:00
# We eat spacers AND any other character, so the following is not needed,
# although the lex file includes it (the lex file does not count() invalid characters
# for the purpose of error reporting).
#if c in ' \n\r\x0B':
# continue
except EInternal:
pass # clear the exception and fall through
return ('EOF',)
def NextToken(self):
"""Calls GetToken and sets the internal token."""
self.tok = self.GetToken()
# Recursive-descendent parser. The result is an AST and a symbol table.
def expect(self, toktype):
"""Raise exception if the current token is not the given one."""
if self.tok[0] != toktype:
if self.tok[0] == 'EOF':
raise EParseUEOF(self)
raise EParseSyntax(self)
def does_something(self, blk):
"""Tell if a list of nodes does something or is just empty statements
(a pure combination of ';' and '{}' and '@')
"""
for node in blk:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if '@' != node.nt != ';':
if node.nt == '{}':
if self.does_something(node.ch):
return True
else:
return True
return False
def Parse_vector_rotation_tail(self):
"""(See Parse_unary_postfix_expression for context)
To our advantage, the precedence of the closing '>' in a vector or
rotation literal is that of an inequality. Our strategy will thus be
to perform the job of an inequality, calling the lower level 'shift'
rule and building the inequalities if they are not '>'. When we find a
'>', we check whether the next token makes sense as beginning an
inequality; if not, we finally close the vector or rotation.
But first, a quaternion _may_ have a full expression at the third
component, so we tentatively parse this position as an expression, and
backtrack if it causes an error.
"""
ret = []
pos = self.pos
errorpos = self.errorpos
tok = self.tok
component3 = False
try:
component3 = self.Parse_expression()
# Checking here for '>' might parse a different grammar, because
# it might allow e.g. <1,2,3==3>; as a vector, which is not valid.
# Not too sure about that, but we're cautious and disable this
# just in case.
#if self.tok[0] == '>':
# return ret
self.expect(',')
self.NextToken()
except EParse: # The errors can be varied, e.g. <0,0,0>-v; raises EParseTypeMismatch
# Backtrack
self.pos = pos
self.errorpos = errorpos
self.tok = tok
# We do this here to prevent a type mismatch above
if component3 is not False:
ret.append(self.autocastcheck(component3, 'float'))
# OK, here we are.
inequality = self.Parse_shift() # shift is the descendant of inequality
while self.tok[0] in ('<', '<=', '>=', '>'):
op = self.tok[0]
self.NextToken()
if op == '>':
# Check if the current token can be a part of a comparison.
# If not, it's a vector/quaternion terminator.
if self.tok[0] not in (
# List adapted from this section of the bison report:
#state 570
#
# 176 expression: expression '>' . expression
# 214 quaternion_initializer: '<' expression ',' expression ',' expression ',' expression '>' .
'IDENT', 'INTEGER_VALUE', 'FLOAT_VALUE', 'STRING_VALUE',
'KEY_VALUE', 'VECTOR_VALUE', 'ROTATION_VALUE', 'LIST_VALUE',
'TRUE', 'FALSE', '++', '--', 'PRINT', '!', '~', '(', '['
):
ret.append(self.autocastcheck(inequality, 'float'))
return ret
# This is basically a copy/paste of the Parse_inequality handler
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ltype = inequality.t
if ltype not in ('integer', 'float'):
raise EParseTypeMismatch(self)
rexpr = self.Parse_shift()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
rtype = rexpr.t
if rtype not in ('integer', 'float'):
raise EParseTypeMismatch(self)
if ltype != rtype:
if rtype == 'float':
inequality = self.autocastcheck(inequality, rtype)
else:
rexpr = self.autocastcheck(rexpr, ltype)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
inequality = nr(nt=op, t='integer', ch=[inequality, rexpr])
# Reaching this means an operator or lower precedence happened,
# e.g. <1,1,1,2==2> (that's syntax error in ==)
raise EParseSyntax(self)
def Parse_unary_postfix_expression(self, AllowAssignment = True):
"""Grammar parsed here:
unary_postfix_expression: TRUE | FALSE | LIST_VALUE
| INTEGER_VALUE | FLOAT_VALUE | '-' INTEGER_VALUE | '-' FLOAT_VALUE
| STRING_VALUE | KEY_VALUE | VECTOR_VALUE | ROTATION_VALUE
| vector_literal | rotation_literal | list_literal
| PRINT '(' expression ')' | IDENT '(' expression_list ')'
| lvalue '++' | lvalue '--' | assignment %if allowed
| IDENT '[' expression ']' '=' expression %if lazylists
| IDENT '[' expression ']' %if lazylists
| lvalue
vector_literal: '<' expression ',' expression ',' expression '>'
rotation_literal: '<' expression ',' expression ',' expression
',' expression '>'
list_literal: '[' optional_expression_list ']'
assignment: lvalue '=' expression | lvalue '+=' expression
| lvalue '-=' expression | lvalue '*=' expression
| lvalue '/=' expression | lvalue '%=' expression
| lvalue '|=' expression %if extendedassignment
| lvalue '&=' expression %if extendedassignment
| lvalue '<<=' expression %if extendedassignment
| lvalue '>>=' expression %if extendedassignment
lvalue: IDENT | IDENT '.' IDENT
"""
tok0 = self.tok[0]
val = self.tok[1] if len(self.tok) > 1 else None
CONST = 'CONST'
if tok0 == '-':
self.NextToken()
if self.tok[0] in ('INTEGER_VALUE', 'FLOAT_VALUE'):
val = self.tok[1]
self.NextToken()
return nr(nt=CONST, value=lslfuncs.neg(val),
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
t='integer' if type(val) == int else 'float')
raise EParseSyntax(self)
if tok0 == 'INTEGER_VALUE':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=CONST, t='integer', value=val)
if tok0 == 'FLOAT_VALUE':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=CONST, t='float', value=val)
if tok0 == 'STRING_VALUE':
self.NextToken()
if self.allowmultistrings:
while self.tok[0] == 'STRING_VALUE':
val += self.tok[1]
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=CONST, t='string', value=val)
# Key constants are not currently supported - use string
#if tok0 == 'KEY_VALUE':
# return [CONST, 'key', val]
if tok0 == 'VECTOR_VALUE':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=CONST, t='vector', value=val)
if tok0 == 'ROTATION_VALUE':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=CONST, t='rotation', value=val)
if tok0 == 'LIST_VALUE':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=CONST, t='list', value=val)
if tok0 in ('TRUE', 'FALSE'):
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=CONST, t='integer', value=1 if tok0 == 'TRUE' else 0)
if tok0 == '<':
self.NextToken()
saveAllowVoid = self.allowVoid
self.allowVoid = False
val = [self.autocastcheck(self.Parse_expression(), 'float')]
self.expect(',')
self.NextToken()
val.append(self.autocastcheck(self.Parse_expression(), 'float'))
self.expect(',')
self.NextToken()
# It would be cute if it were this simple:
#val.append(self.Parse_expression())
#if self.tok[0] == '>':
# self.NextToken()
# return ['VECTOR', 'vector'] + val
#self.expect(',')
#self.NextToken()
#val.append(self.Parse_inequality())
#self.expect('>')
#self.NextToken()
#return ['ROTATION', 'rotation'] + val
# Alas, it isn't. The closing angle bracket of a vector '>'
# conflicts with the inequality operator '>' in unexpected ways.
# Example: <2,2,2> * 2 would trigger the problem with that code:
# the expression parser would try to parse the inequality 2 > *2,
# choking at the *. To make things worse, LSL admits things such as
# <2,2,2 > 2> (but not things like <2,2,2 == 2> because the == has
# lower precedence than the '>' and thus it forces termination of
# the vector constant). And to make things even worse, it also
# admits things such as <2,2,2 == 2, 2> because the comma is not in
# the precedence scale, so it's quite complex to handle.
# We defer it to a separate function.
val += self.Parse_vector_rotation_tail()
self.allowVoid = saveAllowVoid
if len(val) == 3:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='VECTOR', t='vector', ch=val)
return nr(nt='ROTATION', t='rotation', ch=val)
if tok0 == '[':
self.NextToken()
val = self.Parse_optional_expression_list(False)
self.expect(']')
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='LIST', t='list', ch=val)
if tok0 == 'PRINT':
self.NextToken()
self.expect('(')
self.NextToken()
saveAllowVoid = self.allowVoid
self.allowVoid = True
expr = self.Parse_expression()
self.allowVoid = saveAllowVoid
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if expr.t not in types:
raise (EParseTypeMismatch(self) if expr.t is None
else EParseUndefined(self))
self.expect(')')
self.NextToken()
# Syntactically, print returns the same type as the expression.
# However, compilation in Mono throws an exception, and even in
# LSO, it throws a bounds check error when the result is a string
# or key or list and the returned value is used.
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='PRINT', t=expr.t, ch=[expr])
if tok0 != 'IDENT':
if tok0 == 'EOF':
raise EParseUEOF(self)
raise EParseSyntax(self)
name = val
savepos = self.errorpos
self.NextToken()
# Course of action decided here.
tok0 = self.tok[0]
if tok0 == '(':
# Function call
self.NextToken()
# Functions are looked up in the global scope only.
sym = self.FindSymbolFull(val, globalonly=True)
if sym is None:
self.errorpos = savepos
raise EParseUndefined(self)
if sym['Kind'] != 'f':
self.errorpos = savepos
raise EParseUndefined(self)
args = self.Parse_optional_expression_list(sym['ParamTypes'])
self.expect(')')
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='FNCALL', t=sym['Type'], name=name, ch=args)
sym = self.FindSymbolFull(val)
if sym is None or sym['Kind'] not in {'v', 'c'}:
self.errorpos = savepos
raise EParseUndefined(self)
typ = sym['Type']
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
lvalue = nr(nt='IDENT', t=typ, name=name, scope=sym['Scope'])
# Lazy lists
if self.lazylists and tok0 == '[':
self.NextToken()
if typ != 'list':
raise EParseTypeMismatch(self)
idxexpr = self.Parse_optional_expression_list(False)
self.expect(']')
self.NextToken()
if self.tok[0] != '=' or not AllowAssignment:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='SUBIDX', t=None, ch=[lvalue] + idxexpr)
# Lazy list assignment
if len(idxexpr) != 1:
raise EParseFunctionMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if idxexpr[0].t != 'integer':
raise EParseTypeMismatch(self)
idxexpr = idxexpr[0]
self.NextToken()
saveAllowVoid = self.allowVoid
self.allowVoid = True
expr = self.Parse_expression()
self.allowVoid = saveAllowVoid
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
rtyp = expr.t
# Define aux function if it doesn't exist
# (leaves users room for writing their own replacement, e.g.
# one that uses something other than integer zero as filler)
if 'lazy_list_set' not in self.symtab[0]:
self.PushScope()
paramscope = self.scopeindex
self.PushScope()
blockscope = self.scopeindex
params = (['list', 'integer', 'list'],
['L', 'i', 'v'])
self.AddSymbol('f', 0, 'lazy_list_set', Loc=self.usedspots,
Type='list', ParamTypes=params[0], ParamNames=params[1],
Inline=False)
for typ, name in zip(*params):
self.AddSymbol('v', paramscope, name, Type=typ)
#self.PushScope() # no locals
# Add body (apologies for the wall of text)
# Generated from this source:
'''
list lazy_list_set(list L, integer i, list v)
{
while (llGetListLength(L) < i)
L = L + 0;
return llListReplaceList(L, v, i, i);
}
'''
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
self.tree[self.usedspots] = nr(
nt='FNDEF'
, t='list'
, name='lazy_list_set'
, ptypes=params[0]
, pnames=params[1]
, scope=0
, pscope=paramscope
, ch=[
nr(nt='{}'
, t=None
, LIR=True
, scope=blockscope
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
, ch=[
nr(nt='WHILE'
, t=None
, ch=[
nr(nt='<'
, t='integer'
, ch=[
nr(nt='FNCALL'
, t='integer'
, name='llGetListLength'
, ch=[
nr(nt='IDENT'
, t='list'
, name='L'
, scope=paramscope
)
]
),
nr(nt='IDENT'
, t='integer'
, name='i'
, scope=paramscope
)
]
),
nr(nt='EXPR'
, t='list'
, ch=[
nr(nt='='
, t='list'
, ch=[
nr(nt='IDENT'
, t='list'
, name='L'
, scope=paramscope
),
nr(nt='+'
, t='list'
, ch=[
nr(nt='IDENT'
, t='list'
, name='L'
, scope=paramscope
),
nr(nt='CONST'
, t='integer'
, value=0
)
]
)
]
)
]
)
]
),
nr(nt='RETURN'
, t=None
, LIR=True
, ch=[
nr(nt='FNCALL'
, t='list'
, name='llListReplaceList'
, ch=[
nr(nt='IDENT'
, t='list'
, name='L'
, scope=paramscope
),
nr(nt='IDENT'
, t='list'
, name='v'
, scope=paramscope
),
nr(nt='IDENT'
, t='integer'
, name='i'
, scope=paramscope
),
nr(nt='IDENT'
, t='integer'
, name='i'
, scope=paramscope
)
]
)
]
)
]
)
]
)
self.usedspots += 1
self.PopScope()
self.PopScope()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if expr.t is None:
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if expr.t != 'list':
expr = nr(nt='CAST', t='list', ch=[expr])
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='=', t='list', ch=[lvalue, nr(
nt='FNCALL', t='list', name='lazy_list_set', scope=0,
ch=[lvalue.copy(), idxexpr, expr]
)])
if tok0 == '.':
if sym['Kind'] == 'c':
raise EParseSyntax(self)
self.NextToken()
self.expect('IDENT')
self.ValidateField(typ, self.tok[1])
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
lvalue = nr(nt='FLD', t='float', ch=[lvalue], fld=self.tok[1])
self.NextToken()
tok0 = self.tok[0]
typ = 'float'
if tok0 in ('++', '--'):
if sym['Kind'] == 'c':
raise EParseSyntax(self)
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if lvalue.t not in ('integer', 'float'):
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='V++' if tok0 == '++' else 'V--', t=lvalue.t,
ch=[lvalue])
if AllowAssignment and (tok0 in self.assignment_toks
or self.extendedassignment
and tok0 in self.extassignment_toks):
if sym['Kind'] == 'c':
raise EParseSyntax(self)
self.NextToken()
expr = self.Parse_expression()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
rtyp = expr.t
if typ in ('integer', 'float'):
# LSL admits integer *= float (go figger).
# It acts like: lhs = (integer)((float)lhs * rhs)
# That would trigger an error without this check.
if tok0 != '*=' or typ == 'float':
expr = self.autocastcheck(expr, typ)
rtyp = typ
# Lots of drama for checking types. This is pretty much like
# addition, subtraction, multiply, divide, etc. all in one go.
if tok0 == '=':
expr = self.autocastcheck(expr, typ)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='=', t=typ, ch=[lvalue, expr])
if tok0 == '+=':
if typ == 'float':
expr = self.autocastcheck(expr, typ)
if rtyp != typ != 'list' or typ == rtyp == 'key':
# key + key is the only disallowed combo of equal types
raise EParseTypeMismatch(self)
if self.explicitcast:
if typ == 'list' != rtyp:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
expr = nr(nt='CAST', t=typ, ch=[expr])
return nr(nt=tok0, t=typ, ch=[lvalue, expr])
if tok0 == '-=':
if typ == rtyp in ('integer', 'float', 'vector', 'rotation'):
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=tok0, t=typ, ch=[lvalue, expr])
raise EParseTypeMismatch(self)
if tok0 in ('*=', '/='):
2015-03-09 05:20:37 -07:00
# There is a special case that was dealt with before.
if tok0 == '*=' and typ == 'integer' and rtyp == 'float':
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=tok0, t=typ, ch=[lvalue, expr])
if (typ == rtyp or typ == 'vector') and rtyp in ('integer', 'float', 'rotation'):
if typ == 'vector' and rtyp == 'integer':
expr = self.autocastcheck(expr, 'float')
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=tok0, t=typ, ch=[lvalue, expr])
raise EParseTypeMismatch(self)
if tok0 == '%=':
if typ == rtyp in ('integer', 'vector'):
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=tok0, t=typ, ch=[lvalue, expr])
raise EParseTypeMismatch(self)
# Rest take integer operands only
if typ == rtyp == 'integer':
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=tok0, t=typ, ch=[lvalue, expr])
raise EParseTypeMismatch(self)
return lvalue
def Parse_unary_expression(self, AllowAssignment = True):
"""Grammar parsed here:
unary_expression: '-' factor | '!' unary_expression | '~' unary_expression
# we expand lvalue here to facilitate parsing
| '++' IDENT | '++' IDENT '.' IDENT
| '--' IDENT | '--' IDENT '.' IDENT
| '(' TYPE ')' typecast_expression | '(' expression ')'
| unary_postfix_expression
%NORMAL RULES ONLY:
typecast_expression: '(' expression ')'
| unary_postfix_expression %except assignment
%EXTENDED RULES ONLY:
typecast_expression: unary_expression %except assignment
"""
tok0 = self.tok[0]
if tok0 == '-':
# Unary minus
self.NextToken()
value = self.Parse_factor()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if value.t not in ('integer', 'float', 'vector', 'rotation'):
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='NEG', t=value.t, ch=[value])
if tok0 in ('!', '~'):
# Unary logic and bitwise NOT - applies to integers only
self.NextToken()
value = self.Parse_unary_expression()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if value.t != 'integer':
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=tok0, t='integer', ch=[value])
if tok0 in ('++', '--'):
# Pre-increment / pre-decrement
self.NextToken()
self.expect('IDENT')
name = self.tok[1]
sym = self.FindSymbolFull(name)
if sym is None or sym['Kind'] != 'v':
# Pretend it doesn't exist
raise EParseUndefined(self)
typ = sym['Type']
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret = nr(nt='IDENT', t=typ, name=name, scope=sym['Scope'])
self.NextToken()
if self.tok[0] == '.':
self.NextToken()
self.expect('IDENT')
self.ValidateField(typ, self.tok[1])
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret = nr(nt='FLD', t='float', ch=[ret], fld=self.tok[1])
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
typ = ret.t
if typ not in ('integer', 'float'):
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='++V' if tok0 == '++' else '--V', t=typ, ch=[ret])
if tok0 == '(':
# Parenthesized expression or typecast
self.NextToken()
if self.tok[0] != 'TYPE':
# Parenthesized expression
expr = self.Parse_expression()
self.expect(')')
self.NextToken()
return expr
# Typecast
typ = self.tok[1]
self.NextToken()
self.expect(')')
self.NextToken()
if self.extendedtypecast:
# Allow any unary expression (except assignment). The type cast
# acts as a prefix operator.
# Deal with the case of minus a constant integer or float.
# E.g. ~(integer)-2*3 should be parsed as (~(integer)-2)*3
# and not as ~(integer)(-(2*3))
# Note ~(integer)-a*3 is also parsed as ~(integer)(-a)*3
# which is bordering a violation of the POLA because of the
# priority of - with respect to *. But the syntax is quite
# explicit: what is typecast is always a unary expression,
# therefore processed first.
if self.tok[0] == '-':
self.NextToken()
if self.tok[0] == 'INTEGER_VALUE':
expr = nr(nt='CONST', t='integer',
value=lslfuncs.neg(self.tok[1]))
self.NextToken()
elif self.tok[0] == 'FLOAT_VALUE':
expr = nr(nt='CONST', t='float',
value=lslfuncs.neg(self.tok[1]))
self.NextToken()
else:
expr = self.Parse_unary_expression(AllowAssignment = False)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
expr = nr(nt='NEG', t=expr.t, ch=[expr])
else:
expr = self.Parse_unary_expression(AllowAssignment = False)
else:
if self.tok[0] == '(':
self.NextToken()
expr = self.Parse_expression()
self.expect(')')
self.NextToken()
else:
expr = self.Parse_unary_postfix_expression(AllowAssignment = False)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
basetype = expr.t
if self.lazylists and basetype is None and expr.nt == 'SUBIDX':
if typ not in self.TypeToExtractionFunction:
raise EParseNoConversion(self)
fn = self.TypeToExtractionFunction[typ]
sym = self.FindSymbolFull(fn, globalonly=True)
assert sym is not None
fnparamtypes = sym['ParamTypes']
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
subparamtypes = [x.t for x in expr.ch]
if fnparamtypes != subparamtypes:
raise EParseFunctionMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='FNCALL', t=sym['Type'], name=fn, scope=0,
ch=expr.ch)
if typ == 'list' and basetype in types \
or basetype in ('integer', 'float') and typ in ('integer', 'float', 'string') \
or basetype == 'string' and typ in types \
or basetype == 'key' and typ in ('string', 'key') \
or basetype == 'vector' and typ in ('string', 'vector') \
or basetype == 'rotation' and typ in ('string', 'rotation') \
or basetype == 'list' and typ == 'string':
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='CAST', t=typ, ch=[expr])
raise EParseTypeMismatch(self)
# Must be a postfix expression.
return self.Parse_unary_postfix_expression(AllowAssignment)
def Parse_factor(self):
"""Grammar parsed here:
factor: unary_expression | factor '*' unary_expression
| factor '/' unary_expresssion | factor '%' unary_expression
"""
factor = self.Parse_unary_expression()
while self.tok[0] in ('*', '/', '%'):
op = self.tok[0]
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ltype = factor.t
# Acceptable types for LHS
if op in ('*', '/') and ltype not in ('integer', 'float',
'vector', 'rotation') \
or op == '%' and ltype not in ('integer', 'vector'):
raise EParseTypeMismatch(self)
self.NextToken()
rexpr = self.Parse_unary_expression()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
rtype = rexpr.t
# Mod is easier to check for
if op == '%' and ltype != rtype:
raise EParseTypeMismatch(self)
if op == '%' or ltype == rtype == 'integer':
# Deal with the special cases first (it's easy)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
factor = nr(nt=op, t=ltype, ch=[factor, rexpr])
else:
# Any integer must be promoted to float now
if ltype == 'integer':
ltype = 'float'
factor = self.autocastcheck(factor, ltype)
if rtype == 'integer':
rtype = 'float'
rexpr = self.autocastcheck(rexpr, rtype)
if ltype == 'float' and rtype in ('float', 'vector') \
or ltype == 'vector' and rtype in ('float', 'vector', 'rotation') \
or ltype == rtype == 'rotation':
if op == '/' and rtype == 'vector':
# Division by vector isn't valid
raise EParseTypeMismatch(self)
# The rest are valid
if ltype == 'float' and rtype == 'vector':
resulttype = rtype
elif ltype == rtype == 'vector':
resulttype = 'float'
else:
resulttype = ltype
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
factor = nr(nt=op, t=resulttype, ch=[factor, rexpr])
else:
raise EParseTypeMismatch(self)
return factor
def Parse_term(self):
"""Grammar parsed here:
term: factor | term '+' factor | term '-' factor
"""
term = self.Parse_factor()
while self.tok[0] in ('+', '-'):
op = self.tok[0]
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ltype = term.t
if op == '+' and ltype not in types \
or op == '-' and ltype not in ('integer', 'float',
'vector', 'rotation'):
raise EParseTypeMismatch(self)
self.NextToken()
rexpr = self.Parse_factor()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
rtype = rexpr.t
# This is necessary, but the reason is subtle.
# The types must match in principle (except integer/float), so it
# doesn't seem necessary to check rtype. But there's the case
# where the first element is a list, where the types don't need to
# match but the second type must make sense.
if op == '+' and rtype not in types:
#or op == '-' and rtype not in ('integer', 'float',
# 'vector', 'rotation'):
raise EParseTypeMismatch(self)
# Isolate the additions where the types match to make our life easier later
if op == '+' and (ltype == rtype or ltype == 'list' or rtype == 'list'):
if ltype == rtype == 'key':
# key + key is the only disallowed combo of equals
raise EParseTypeMismatch(self)
# Note that although list + nonlist is semantically the
# same as list + (list)nonlist, and similarly for
# nonlist + list, they don't compile to the same thing,
# so we don't act on self.explicitcast in this case.
if rtype == 'list':
ltype = rtype
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
term = nr(nt=op, t=ltype, ch=[term, rexpr])
elif self.allowkeyconcat and op == '+' \
and ltype in ('key', 'string') and rtype in ('key', 'string'):
2014-07-25 19:44:02 -07:00
# Allow string+key addition (but add explicit cast)
if ltype == 'key':
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
term = nr(nt=op, t=rtype,
ch=[nr(nt='CAST', t=rtype, ch=[term]), rexpr])
else:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
term = nr(nt=op, t=ltype,
ch=[term, nr(nt='CAST', t=ltype, ch=[rexpr])])
elif ltype == 'key' or rtype == 'key':
# Only list + key or key + list is allowed, otherwise keys can't
# be added or subtracted with anything.
raise EParseTypeMismatch(self)
else:
if ltype == 'float':
# Promote rexpr to float
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
term = nr(nt=op, t=ltype,
ch=[term, self.autocastcheck(rexpr, ltype)])
else:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
# Convert LHS to rtype if possible (note no keys get here)
term = nr(nt=op, t=rtype,
ch=[self.autocastcheck(term, rtype), rexpr])
return term
def Parse_shift(self):
"""Grammar parsed here:
shift: term | shift '<<' term | shift '>>' term
"""
shift = self.Parse_term()
while self.tok[0] in ('<<', '>>'):
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if shift.t != 'integer':
raise EParseTypeMismatch(self)
op = self.tok[0]
self.NextToken()
rexpr = self.Parse_term()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if rexpr.t != 'integer':
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
shift = nr(nt=op, t='integer', ch=[shift , rexpr])
return shift
def Parse_inequality(self):
"""Grammar parsed here:
inequality: shift | inequality '<' shift | inequality '<=' shift
| inequality '>' shift | inequality '>=' shift
"""
inequality = self.Parse_shift()
while self.tok[0] in ('<', '<=', '>', '>='):
op = self.tok[0]
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ltype = inequality.t
if ltype not in ('integer', 'float'):
raise EParseTypeMismatch(self)
self.NextToken()
rexpr = self.Parse_shift()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
rtype = rexpr.t
if rtype not in ('integer', 'float'):
raise EParseTypeMismatch(self)
if ltype != rtype:
if rtype == 'float':
inequality = self.autocastcheck(inequality, rtype)
else:
rexpr = self.autocastcheck(rexpr, ltype)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
inequality = nr(nt=op, t='integer', ch=[inequality, rexpr])
return inequality
def Parse_comparison(self):
"""Grammar parsed here:
comparison: inequality | comparison '==' inequality
| comparison '!=' inequality
"""
comparison = self.Parse_inequality()
while self.tok[0] in ('==', '!='):
op = self.tok[0]
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ltype = comparison.t
if ltype not in types:
raise EParseTypeMismatch(self)
self.NextToken()
rexpr = self.Parse_inequality()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
rtype = rexpr.t
if ltype == 'float':
rexpr = self.autocastcheck(rexpr, ltype)
else:
# For string & key, RHS (rtype) mandates the conversion
# (that's room for optimization: always compare strings)
comparison = self.autocastcheck(comparison, rtype)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
comparison = nr(nt=op, t='integer', ch=[comparison, rexpr])
return comparison
def Parse_bitbool_factor(self):
"""Grammar parsed here:
bitbool_factor: comparison | bitbool_factor '&' comparison
"""
bitbool_factor = self.Parse_comparison()
while self.tok[0] == '&':
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if bitbool_factor.t != 'integer':
raise EParseTypeMismatch(self)
op = self.tok[0]
self.NextToken()
rexpr = self.Parse_comparison()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if rexpr.t != 'integer':
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
bitbool_factor = nr(nt=op, t='integer', ch=[bitbool_factor, rexpr])
return bitbool_factor
def Parse_bitxor_term(self):
"""Grammar parsed here:
bitxor_term: bitbool_factor | bitxor_term '^' bitbool_factor
"""
bitxor_term = self.Parse_bitbool_factor()
while self.tok[0] == '^':
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if bitxor_term.t != 'integer':
raise EParseTypeMismatch(self)
op = self.tok[0]
self.NextToken()
rexpr = self.Parse_bitbool_factor()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if rexpr.t != 'integer':
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
bitxor_term = nr(nt=op, t='integer', ch=[bitxor_term, rexpr])
return bitxor_term
def Parse_bitbool_term(self):
"""Grammar parsed here:
bitbool_term: bitxor_term | bitbool_term '|' bitxor_term
"""
bitbool_term = self.Parse_bitxor_term()
while self.tok[0] == '|':
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if bitbool_term.t != 'integer':
raise EParseTypeMismatch(self)
op = self.tok[0]
self.NextToken()
rexpr = self.Parse_bitxor_term()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if rexpr.t != 'integer':
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
bitbool_term = nr(nt=op, t='integer', ch=[bitbool_term, rexpr])
return bitbool_term
def Parse_expression(self):
"""Grammar parsed here:
expression: bitbool_term | expression '||' bitbool_term
| expression '&&' bitbool_term
Most operators with same priority, in general, are executed in
right-to-left order but calculated with precedence left-to-right.
That is, the tree is generated LTR but traversed RTL (in post-order).
E.g. a-b+c is calculated (in RPN notation) as: c, b, a, swap, -, +
i.e. c is evaluated first and a last, but the operation is still (a-b)+c
which is normal LTR.
At this point we're just constructing the tree, so we follow normal
precedence rules.
"""
expression = self.Parse_bitbool_term()
while self.tok[0] in ('&&', '||'):
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if expression.t != 'integer':
raise EParseTypeMismatch(self)
op = self.tok[0]
self.NextToken()
rexpr = self.Parse_bitbool_term()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if rexpr.t != 'integer':
raise EParseTypeMismatch(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
expression = nr(nt=op, t='integer', ch=[expression, rexpr])
if not self.allowVoid and expression.t not in types:
raise EParseTypeMismatch(self)
return expression
def Parse_optional_expression_list(self, expected_types = None):
"""Grammar parsed here:
optional_expression_list: LAMBDA | expression_list
expression_list: expression | expression_list ',' expression
"""
# Recursive descendent parsers are nice, but not exempt of problems.
# We need to accept empty lists. This is a maze of which we get out
# with a dirty hack. Rather than attempt to parse as an expression and
# backtrack in case of error, we check the next token to see if it
# is one that closes the expression list.
# optional_expression_list is used by FOR loops (closed by ';' or ')'),
# list constants and lazy lists (closed by ']') and function arguments
# (closed by ')'). If it's not the right token, we'll err anyway upon
# return.
ret = []
idx = 0
if self.tok[0] not in (']', ')', ';'):
while True:
saveAllowVoid = self.allowVoid
self.allowVoid = True
expr = self.Parse_expression()
self.allowVoid = saveAllowVoid
if expr.nt == 'SUBIDX' and expr.t is None:
# Don't accept an untyped lazy list in expression lists
raise EParseTypeMismatch(self)
if False is not expected_types is not None:
if idx >= len(expected_types):
raise EParseFunctionMismatch(self)
try:
expr = self.autocastcheck(expr, expected_types[idx]);
except EParseTypeMismatch:
raise EParseFunctionMismatch(self)
elif expected_types is False and self.optenabled:
# don't accept void expressions if optimization is on
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if expr.t not in types:
raise EParseTypeMismatch(self)
idx += 1
ret.append(expr)
if self.tok[0] != ',':
break
self.NextToken()
if False is not expected_types is not None and idx != len(expected_types):
raise EParseFunctionMismatch(self)
return ret
def Parse_statement(self, ReturnType, AllowDecl = False, AllowStSw = False,
InsideSwitch = False, InsideLoop = False):
"""Grammar parsed here:
statement: ';' | single_statement | code_block
single_statement: if_statement | while_statement | do_statement
| for_statement | jump_statement | state_statement | label_statement
| return_statement | declaration_statement | expression ';'
| switch_statement %if enableswitch
| case_statement %if enableswitch and InsideSwitch
| break_statement %if enableswitch and InsideSwitch or breakcont and InsideLoop
| continue_statement %if breakcont and InsideLoop
if_statement: IF '(' expression ')' statement ELSE statement
| IF '(' expression ')' statement
while_statement: WHILE '(' expression ')' statement
do_statement: DO statement WHILE '(' expression ')' ';'
for_statement: FOR '(' optional_expression_list ';' expression ';'
optional_expression_list ')' statement
jump_statement: JUMP IDENT ';'
state_statement: STATE DEFAULT ';' | STATE IDENT ';'
label_statement: '@' IDENT ';'
return_statement: RETURN ';' | RETURN expression ';'
declaration_statement: TYPE lvalue ';' | TYPE lvalue '=' expression ';'
switch_statement: SWITCH '(' expression ')' code_block
case_statement: CASE expression ':' | CASE expression code_block
| DEFAULT ':' | DEFAULT code_block
break_statement: BREAK ';'
continue_statement: CONTINUE ';'
There's a restriction: a *single* statement can not be a declaration.
For example: if (1) integer x; is not allowed.
Note that SWITCH expects a code block because CASE is a full statement
for us, rather than a label. So for example this wouldn't work:
switch (expr) case expr: stmt; // works in C but not in this processor
but this works in both: switch (expr) { case expr: stmt; }
"""
tok0 = self.tok[0]
if tok0 == '{':
return self.Parse_code_block(ReturnType, AllowStSw = AllowStSw,
InsideSwitch = InsideSwitch, InsideLoop = InsideLoop)
if tok0 == ';':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=';', t=None)
if tok0 == '@':
if not AllowDecl and self.optenabled:
raise EParseInvalidLabelOpt(self)
self.NextToken()
self.expect('IDENT')
name = self.tok[1]
if name in self.symtab[0] and self.symtab[0][name]['Kind'] == 'c':
raise EParseSyntax(self)
if name in self.symtab[self.scopeindex]:
raise EParseAlreadyDefined(self)
# shrinknames *needs* all labels renamed, so they are out of the way
if self.duplabels or self.shrinknames:
# Duplicate labels allowed.
if name in self.locallabels or self.shrinknames:
# Generate a new unique name and attach it to the symbol.
unique = self.GenerateLabel()
self.AddSymbol('l', self.scopeindex, name, NewName=unique,
ref=0)
else:
# Use the existing name. Faster and more readable.
unique = name
self.locallabels.add(name)
self.AddSymbol('l', self.scopeindex, name, ref=0)
else:
# Duplicate labels disallowed.
# All labels go to a common pool local to the current function.
# Check if it's already there, and add it otherwise.
if name in self.locallabels:
raise EParseDuplicateLabel(self)
self.locallabels.add(name)
self.AddSymbol('l', self.scopeindex, name, ref=0)
self.NextToken()
self.expect(';')
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='@', t=None, name=name, scope=self.scopeindex)
if tok0 == 'JUMP':
self.NextToken()
self.expect('IDENT')
name = self.tok[1]
sym = self.FindSymbolPartial(name, MustBeLabel=True)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
jumpnode = nr(nt='JUMP', t=None, name=name, scope=None)
if not sym or sym['Kind'] != 'l':
# It might still be a forward reference, so we add it to the
# list of things to look up when done
self.jump_lookups.append((name, self.scopestack[:],
self.errorpos, jumpnode))
else:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
jumpnode.scope = sym['Scope']
sym['ref'] += 1
self.NextToken()
self.expect(';')
self.NextToken()
return jumpnode
if tok0 == 'STATE':
if self.localevents is None:
if AllowStSw is False:
raise EParseCantChangeState(self)
if AllowStSw is None:
self.PruneBug.append((self.errorpos,
EParseCantChangeState))
self.NextToken()
if self.tok[0] not in ('DEFAULT', 'IDENT'):
raise EParseSyntax(self)
2014-07-31 14:21:50 -07:00
# State Switch only searches for states in the global scope
name = self.tok[1] if self.tok[0] == 'IDENT' else 'default'
if (name not in self.symtab[0]
or self.symtab[0][name]['Kind'] != 's'
) and (name not in self.globals
or self.globals[name]['Kind'] != 's'
):
raise EParseUndefined(self)
self.NextToken()
self.expect(';')
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='STSW', t=None, name=name, scope=0)
if tok0 == 'RETURN':
savepos = self.errorpos
self.NextToken()
if self.tok[0] == ';':
value = None
else:
savepos = self.errorpos
saveAllowVoid = self.allowVoid
# Needed due to another LSL bug, see regr/void-in-return.lsl
self.allowVoid = True
value = self.Parse_expression()
self.allowVoid = saveAllowVoid
self.expect(';')
self.NextToken()
if ReturnType is None and value is not None:
# It follows the same rules as AllowStSw
if AllowStSw is False:
self.errorpos = savepos
raise EParseReturnShouldBeEmpty(self)
elif value.t is None:
if AllowStSw is None:
self.PruneBug.append((self.errorpos,
EParseReturnShouldBeEmpty))
self.PushScope()
scope = self.scopeindex
self.PopScope()
return nr(nt='{}', t=None, scope=scope,
ch=[nr(nt='EXPR', t=None, ch=[value]),
nr(nt='RETURN', t=None)])
else:
self.errorpos = savepos
raise EParseTypeMismatch(self)
if ReturnType is not None and value is None:
self.errorpos = savepos
raise EParseReturnIsEmpty(self)
if value is None:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='RETURN', t=None)
# Sets LastIsReturn flag too
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='RETURN', t=None, LIR=True,
ch=[self.autocastcheck(value, ReturnType)])
if tok0 == 'IF':
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret = nr(nt='IF', t=None, ch=[])
self.NextToken()
self.expect('(')
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret.ch.append(self.Parse_expression())
self.expect(')')
self.NextToken()
savePruneBug = self.PruneBug
self.PruneBug = []
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret.ch.append(self.Parse_statement(ReturnType, AllowStSw = None, InsideLoop = InsideLoop))
if self.tok[0] == 'ELSE':
if AllowStSw is False and self.PruneBug:
self.errorpos = self.PruneBug[0][0]
raise self.PruneBug[0][1](self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
LastIsReturn = getattr(ret.ch[1], 'LIR', False)
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret.ch.append(self.Parse_statement(ReturnType,
AllowStSw = AllowStSw, InsideLoop = InsideLoop))
if AllowStSw is None:
savePruneBug += self.PruneBug
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if LastIsReturn and getattr(ret.ch[2], 'LIR', False):
ret.LIR = True
self.PruneBug = savePruneBug
return ret
if tok0 == 'WHILE':
self.NextToken()
if self.breakcont:
# We may add braces - or not. The safe approach is to assume
# we always do and open a new scope for it. At worst it will be
2015-03-05 12:18:13 -07:00
# empty. At least it is not reflected as braces in the code if
# braces are not used.
#
# This is designed to deal with cases like:
# if (a) while (b) { ... break; }
#
# This works by adding braces around the while and the newly
# added label, like this:
# if (a) { while (b) { ... jump label; } @label; }
self.PushScope()
self.breakstack.append([self.GenerateLabel(), self.scopeindex,
0])
# Scope still unknown; if a block is opened, Parse_code_block()
# will fill it in.
self.continuestack.append([self.GenerateLabel(), None, 0])
self.expect('(')
self.NextToken()
condition = self.Parse_expression()
self.expect(')')
self.NextToken()
# To fix a problem with a corner case (LSL allows defining a label
# in a single statement, at the same scope as the loop, breaking
# some of our logic), we check if the statement is a label. If so,
# we pop the scope to parse the statement and push it again.
# It won't cause scope problems in turn because we won't add any
# break or continue labels if no break or continue statement is
# present, which it can't because the statement is a label.
if self.breakcont and self.tok[0] == '@':
self.PopScope()
stmt = self.Parse_statement(ReturnType, AllowStSw = True, InsideLoop = True)
self.PushScope()
else:
stmt = self.Parse_statement(ReturnType, AllowStSw = True, InsideLoop = True)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret = nr(nt='WHILE', t=None, ch=[condition, stmt])
if self.breakcont:
last = self.continuestack.pop()
if last[2]:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
assert ret.ch[1].nt == '{}'
ret.ch[1].ch.append(nr(nt='@', t=None, name=last[0],
scope=last[1]))
self.AddSymbol('l', last[1], last[0], ref=last[2])
last = self.breakstack.pop()
if last[2]:
assert last[1] is not None
ret = nr(nt='{}', t=None, scope=last[1], ch=[ret,
nr(nt='@', t=None, name=last[0], scope=last[1])])
self.AddSymbol('l', last[1], last[0], ref=last[2])
self.PopScope()
return ret
if tok0 == 'DO':
self.NextToken()
if self.breakcont:
self.PushScope()
self.breakstack.append([self.GenerateLabel(), self.scopeindex,
0])
# Scope still unknown; if a block is opened, Parse_code_block()
# will fill it in.
self.continuestack.append([self.GenerateLabel(), None, 0])
if self.breakcont and self.tok[0] == '@':
self.PopScope()
stmt = self.Parse_statement(ReturnType, AllowStSw = True,
InsideLoop = True)
self.PushScope()
else:
stmt = self.Parse_statement(ReturnType, AllowStSw = True,
InsideLoop = True)
self.expect('WHILE')
self.NextToken()
self.expect('(')
self.NextToken()
condition = self.Parse_expression()
self.expect(')')
self.NextToken()
self.expect(';')
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret = nr(nt='DO', t=None, ch=[stmt, condition])
if self.breakcont:
last = self.continuestack.pop()
if last[2]:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
assert ret.ch[0].nt == '{}'
ret.ch[0].ch.append(nr(nt='@', t=None, name=last[0],
scope=last[1]))
self.AddSymbol('l', last[1], last[0], ref=last[2])
last = self.breakstack.pop()
if last[2]:
assert last[1] is not None
ret = nr(nt='{}', t=None, scope=last[1], ch=[ret,
nr(nt='@', t=None, name=last[0], scope=last[1])])
self.AddSymbol('l', last[1], last[0], ref=last[2])
self.PopScope()
return ret
if tok0 == 'FOR':
self.NextToken()
if self.breakcont:
self.PushScope()
self.breakstack.append([self.GenerateLabel(), self.scopeindex,
0])
# Scope still unknown; if a block is opened, Parse_code_block()
# will fill it in.
self.continuestack.append([self.GenerateLabel(), None, 0])
self.expect('(')
self.NextToken()
initializer = self.Parse_optional_expression_list()
self.expect(';')
self.NextToken()
condition = self.Parse_expression()
self.expect(';')
self.NextToken()
iterator = self.Parse_optional_expression_list()
self.expect(')')
self.NextToken()
if self.breakcont and self.tok[0] == '@':
self.PopScope()
stmt = self.Parse_statement(ReturnType, AllowStSw = True,
InsideLoop = True)
self.PushScope()
else:
stmt = self.Parse_statement(ReturnType, AllowStSw = True,
InsideLoop = True)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret = nr(nt='FOR', t=None,
ch=[nr(nt='EXPRLIST', t=None, ch=initializer),
condition,
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
nr(nt='EXPRLIST', t=None, ch=iterator),
stmt
])
if self.breakcont:
last = self.continuestack.pop()
if last[2]:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
assert ret.ch[3].nt == '{}'
ret.ch[3].ch.append(nr(nt='@', t=None, name=last[0],
scope=last[1]))
self.AddSymbol('l', last[1], last[0], ref=last[2])
last = self.breakstack.pop()
if last[2]:
assert last[1] is not None
ret = nr(nt='{}', t=None, scope=last[1], ch=[ret,
nr(nt='@', t=None, name=last[0], scope=last[1])])
self.AddSymbol('l', last[1], last[0], ref=last[2])
self.PopScope()
return ret
if tok0 == 'SWITCH':
self.NextToken()
self.expect('(')
self.NextToken()
expr = self.Parse_expression()
self.expect(')')
self.NextToken()
brk = self.GenerateLabel()
# Scope is determined in Parse_code_block()
self.breakstack.append([brk, None, 0])
blk = self.Parse_code_block(ReturnType, AllowStSw = AllowStSw,
InsideSwitch = True, InsideLoop = InsideLoop)
blkscope = self.breakstack[-1][1]
# Replace the block
# switch (expr1) { case expr2: stmts1; break; default: stmts2; }
# is translated to:
# {
# if (expr1==expr2) jump label1;
# jump labeldef;
#
# @label1;
# stmts1;
# jump labelbrk;
# @labeldef;
# stmts2;
# @labelbrk;
# }
# The prelude is the ifs and the jumps.
# The block gets the cases replaced with labels,
# and the breaks replaced with jumps.
switchcaselist = []
switchcasedefault = None
# Since label scope rules prevent us from being able to jump inside
# a nested block, only one nesting level is considered.
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
assert blk.nt == '{}'
blk = blk.ch # Disregard the '{}' - we'll add it back later
for idx in xrange(len(blk)):
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if blk[idx].nt == 'CASE':
lbl = self.GenerateLabel()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
switchcaselist.append((lbl, blk[idx].ch[0]))
self.AddSymbol('l', blkscope, lbl, ref=0)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
blk[idx] = nr(nt='@', t=None, name=lbl, scope=blkscope)
elif blk[idx].nt == 'DEFAULTCASE':
if switchcasedefault is not None:
raise EParseManyDefaults(self)
lbl = self.GenerateLabel()
switchcasedefault = lbl
self.AddSymbol('l', blkscope, lbl, ref=0)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
blk[idx] = nr(nt='@', name=lbl, scope=blkscope)
prelude = []
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ltype = expr.t
for case in switchcaselist:
rexpr = case[1]
lexpr = expr
if ltype == 'float':
rexpr = self.autocastcheck(rexpr, ltype)
else:
# For string & key, RHS (rtype) mandates the conversion
# (that's room for optimization: always compare strings)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
lexpr = self.autocastcheck(lexpr, rexpr.t)
prelude.append(nr(nt='IF', t=None, ch=[
nr(nt='==', t='integer', ch=[lexpr, rexpr]),
nr(nt='JUMP', t=None, name=case[0], scope=blkscope)
]))
self.symtab[blkscope][case[0]]['ref'] += 1
if switchcasedefault is None:
if self.errmissingdefault:
raise EParseMissingDefault(self)
# Check if it's worth adding a break label. If there's no
# executable code, there's no point. However, this check is
# insufficient. It misses SEF expressions. For that reason,
# this is best left up to a later optimizer that knows about
# SEF. But we do a preliminary elimination here.
if self.does_something(blk):
switchcasedefault = brk
else:
# Check if no code up to the default label does anything.
# If so, remove the label and don't generate the jump.
for i in xrange(len(blk)):
node = blk[i]
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if (node.nt == '@' and node.name == switchcasedefault
and node.scope == blkscope):
switchcasedefault = None
del blk[i]
break
if self.does_something([node]):
break
del i, node
if switchcasedefault is not None:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
prelude.append(nr(nt='JUMP', t=None, name=switchcasedefault,
scope=blkscope))
if switchcasedefault == brk:
# add a reference to it in the break stack
self.breakstack[-1][2] += 1
else:
self.symtab[blkscope][switchcasedefault]['ref'] += 1
last = self.breakstack.pop()
if last[2]:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
blk.append(nr(nt='@', name=brk, scope=blkscope))
self.AddSymbol('l', blkscope, brk, ref=last[2])
return nr(nt='{}', t=None, scope=blkscope, ch=prelude + blk)
if tok0 == 'CASE':
if not InsideSwitch:
raise EParseInvalidCase(self, u"case")
if self.scopeindex != self.breakstack[-1][1]:
# If this block is nested and not the main switch block, this
# won't work. LSL label scope rules don't expose the nested
# labels. Nothing we can do about that.
raise EParseCaseNotAllowed(self, u"case")
self.NextToken()
expr = self.Parse_expression()
if self.tok[0] == ':':
self.NextToken()
elif self.tok[0] != '{':
raise EParseSyntax(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='CASE', t=None, ch=[expr])
if tok0 == 'DEFAULT':
if self.enableswitch:
if not InsideSwitch:
raise EParseInvalidCase(self, u"default")
if self.scopeindex != self.breakstack[-1][1]:
# If this block is nested and not the main switch block, this
# won't work. Label scope rules don't expose the nested
# labels. Nothing we can do about that.
raise EParseCaseNotAllowed(self, u"default")
self.NextToken()
if self.tok[0] == ':':
self.NextToken()
elif self.tok[0] != '{':
raise EParseSyntax(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='DEFAULTCASE', t=None)
# else fall through to eventually fail
if tok0 == 'BREAK':
if not self.breakstack:
raise EParseInvalidBreak(self)
self.NextToken()
n = -1
if self.tok[0] == 'INTEGER_VALUE':
if self.tok[1] <= 0:
raise EParseInvalidBrkContArg(self)
n = -self.tok[1]
self.NextToken()
self.expect(';')
self.NextToken()
try:
self.breakstack[n][2] += 1
except IndexError:
raise EParseInvalidBrkContArg(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='JUMP', t=None, name=self.breakstack[n][0],
scope=self.breakstack[n][1])
if tok0 == 'CONTINUE':
if not self.continuestack:
raise EParseInvalidCont(self)
self.NextToken()
n = -1
if self.tok[0] == 'INTEGER_VALUE':
if self.tok[1] <= 0:
raise EParseInvalidBrkContArg(self)
n = -self.tok[1]
self.NextToken()
self.expect(';')
self.NextToken()
if n == -1 and self.continuestack[-1][1] is None:
# We're not inside a block - 'continue' is essentially a nop
# e.g. while (cond) continue; is the same as while (cond) ;
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt=';', t='None')
try:
if self.continuestack[n][1] is None:
# this can happen with e.g.:
# while (cond) while (cond) while (cond) continue 3;
# Transform to while(cond) while(cond) while(cond) break 2;
# which is equivalent since there are no {}.
n += 1 # e.g. -3 -> -2
self.breakstack[n][2] += 1 # add a reference to the break
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='JUMP', t=None, name=self.breakstack[n][0],
scope=self.breakstack[n][1])
except IndexError:
raise EParseInvalidBrkContArg(self)
self.continuestack[n][2] += 1
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='JUMP', t=None, name=self.continuestack[n][0],
scope=self.continuestack[n][1])
if tok0 == 'TYPE':
if not AllowDecl:
raise EParseDeclarationScope(self)
typ = self.tok[1]
self.NextToken()
self.expect('IDENT')
name = self.tok[1]
if name in self.symtab[0] and self.symtab[0][name]['Kind'] == 'c':
raise EParseSyntax(self)
if name in self.symtab[self.scopeindex]:
raise EParseAlreadyDefined(self)
self.NextToken()
value = None
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
decl = nr(nt='DECL', t=typ, name=name, scope=self.scopeindex)
if self.tok[0] == '=':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
decl.ch = [self.autocastcheck(self.Parse_expression(), typ)]
self.expect(';')
self.NextToken()
self.AddSymbol('v', self.scopeindex, name, Type=typ)
return decl
# If none of the above, it must be an expression.
saveAllowVoid = self.allowVoid
self.allowVoid = True
value = self.Parse_expression()
self.allowVoid = saveAllowVoid
self.expect(';')
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='EXPR', t=value.t, ch=[value])
def Parse_code_block(self, ReturnType, AllowStSw = False, InsideSwitch = False,
InsideLoop = False):
"""Grammar parsed here:
code_block: '{' statements '}'
statements: LAMBDA | statements statement
It receives the return type to expect for return statements.
"""
self.expect('{')
self.NextToken()
self.PushScope()
# Kludge to find the scope of the break (for switch) /
# continue (for loops) labels.
if self.breakstack: # non-empty iff inside loop or switch
if InsideSwitch and self.breakstack[-1][1] is None:
self.breakstack[-1][1] = self.scopeindex
if InsideLoop and self.continuestack[-1][1] is None:
self.continuestack[-1][1] = self.scopeindex
body = []
LastIsReturn = False
while True:
if self.tok[0] == '}':
self.closebrace = self.errorpos
break
stmt = self.Parse_statement(ReturnType, AllowDecl = True,
AllowStSw = AllowStSw, InsideSwitch = InsideSwitch,
InsideLoop = InsideLoop)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
LastIsReturn = getattr(stmt, 'LIR', False)
body.append(stmt)
scope_braces = self.scopeindex
self.PopScope()
self.expect('}')
self.NextToken()
node = nr(nt='{}', t=None, scope=scope_braces, ch=body)
if LastIsReturn:
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
node.LIR = True
return node
def Parse_simple_expr(self, ForbidList=False):
"""Grammar parsed here:
simple_expr: simple_expr_except_list | list_simple_expr
simple_expr_except_list: STRING_VALUE | KEY_VALUE | VECTOR_VALUE
| ROTATION_VALUE | TRUE | FALSE | IDENT | number_value
| '<' simple_expr ',' simple_expr ',' simple_expr '>'
| '<' simple_expr ',' simple_expr ',' simple_expr ',' simple_expr '>'
number_value: FLOAT_VALUE | INTEGER_VALUE | '-' FLOAT_VALUE | '-' INTEGER_VALUE
list_simple_expr: '[' ']' | '[' list_simple_expr_items ']'
list_simple_expr_items: simple_expr_except_list
| list_simple_expr_items ',' simple_expr_except_list
"""
tok = self.tok
self.NextToken()
if tok[0] in ('TRUE', 'FALSE'): # TRUE and FALSE don't admit sign in globals
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='CONST', t='integer', value=int(tok[0]=='TRUE'))
if tok[0] in ('STRING_VALUE', 'KEY_VALUE', 'VECTOR_VALUE', 'ROTATION_VALUE', 'LIST_VALUE'):
val = tok[1]
if tok[0] == 'STRING_VALUE' and self.allowmultistrings:
while self.tok[0] == 'STRING_VALUE':
val += self.tok[1]
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='CONST', t=lslcommon.PythonType2LSL[type(val)],
value=val)
if tok[0] == 'IDENT':
sym = self.FindSymbolPartial(tok[1])
# The parser accepts library function names here as valid variables
# (it chokes at RAIL in Mono, and at runtime in LSO for some types)
if sym is None or sym['Kind'] not in {'v', 'c'} and (sym['Kind'] !=
'f' or 'ParamNames' in sym): # only UDFs have ParamNames
raise EParseUndefined(self)
typ = sym['Type']
if ForbidList and lslcommon.LSO and typ == 'key':
# This attempts to reproduce LSO's behaviour that a key global
# var inside a list global definition takes a string value
# (SCR-295).
typ = 'string'
return nr(nt='IDENT', t=typ, name=tok[1],
scope=sym['Scope'] if sym['Kind'] == 'v' else 0)
if tok[0] == '<':
value = [self.Parse_simple_expr()]
self.autocastcheck(value[0], 'float')
self.expect(',')
self.NextToken()
value.append(self.Parse_simple_expr())
self.autocastcheck(value[1], 'float')
self.expect(',')
self.NextToken()
value.append(self.Parse_simple_expr())
self.autocastcheck(value[2], 'float')
if self.tok[0] == '>':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='VECTOR', t='vector', ch=value)
self.expect(',')
self.NextToken()
value.append(self.Parse_simple_expr())
self.autocastcheck(value[3], 'float')
self.expect('>')
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='ROTATION', t='rotation', ch=value)
if tok[0] == '[' and not ForbidList:
value = []
if self.tok[0] == ']':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='LIST', t='list', ch=value)
while True:
value.append(self.Parse_simple_expr(ForbidList=True))
if self.tok[0] == ']':
self.NextToken()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='LIST', t='list', ch=value)
self.expect(',')
self.NextToken()
# Integer or Float constant expected
neg = False
if tok[0] == '-':
neg = True
tok = self.tok
self.NextToken()
if tok[0] not in ('INTEGER_VALUE', 'FLOAT_VALUE'):
raise EParseSyntax(self)
value = tok[1]
if neg and (tok[0] != 'INTEGER_VALUE' or value != -2147483648):
value = -value
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
return nr(nt='CONST',
t='float' if tok[0] == 'FLOAT_VALUE' else 'integer', value=value)
def Parse_optional_param_list(self):
"""Grammar parsed here:
optional_param_list: LAMBDA | param_list
param_list: TYPE IDENT | param_list ',' TYPE IDENT
"""
types = []
names = []
if self.tok[0] == 'TYPE':
while True:
typ = self.tok[1]
self.NextToken()
self.expect('IDENT')
name = self.tok[1]
if (name in self.symtab[0]
and self.symtab[0][name]['Kind'] == 'c'
):
raise EParseSyntax(self)
if name in self.symtab[self.scopeindex]:
raise EParseAlreadyDefined(self)
types.append(typ)
names.append(name)
self.AddSymbol('v', self.scopeindex, name, Type=typ, Param=True)
self.NextToken()
if self.tok[0] != ',':
break
self.NextToken()
self.expect('TYPE')
return (types, names)
def Parse_events(self):
"""Grammar parsed here:
events: event | events event
event: EVENT_NAME '(' optional_parameter_list ')' code_block
"""
self.expect('EVENT_NAME') # mandatory
ret = []
while self.tok[0] == 'EVENT_NAME':
name = self.tok[1]
self.NextToken()
if name in self.localevents:
raise EParseAlreadyDefined(self)
self.localevents.add(name)
self.expect('(')
self.NextToken()
# Function parameters go to a dedicated symbol table.
self.PushScope()
params = self.Parse_optional_param_list()
self.expect(')')
self.NextToken()
# NOTE: Parse_events: This is a bit crude, as the error is given at the end of the param list.
# To do it correctly, we can pass the parameter list to Parse_optional_param_list().
if tuple(params[0]) != self.events[name]['pt']:
raise EParseSyntax(self)
self.locallabels = set()
body = self.Parse_code_block(None)
del self.locallabels
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
ret.append(nr(nt='FNDEF', t=None, name=name, # no scope as these are reserved words
pscope=self.scopeindex, ptypes=params[0], pnames=params[1],
ch=[body]))
self.PopScope()
return ret
def Parse_globals(self):
"""Grammar parsed here:
globals: LAMBDA | globals var_def | globals func_def
var_def: TYPE IDENT ';' | TYPE IDENT '=' simple_expr ';'
func_def: optional_type IDENT '(' optional_param_list ')' code_block
optional_type: LAMBDA | TYPE
"""
assert self.scopeindex == 0
while self.tok[0] in ('TYPE','IDENT'):
typ = None
if self.tok[0] == 'TYPE':
typ = self.tok[1]
self.NextToken()
self.expect('IDENT')
name = self.tok[1]
self.NextToken()
if name in self.symtab[0]:
# Duplicate identifier. That's an exception unless function
# override is in effect.
report = True
if self.funcoverride:
# Is it a function definition, and is the entry in the
# symbol table a function definition itself? And is it
# a user-defined function?
if self.tok[0] == '(' \
and self.symtab[0][name]['Kind'] == 'f' \
and 'Loc' in self.symtab[0][name]:
# Override it.
report = False
# Delete the previous definition.
self.tree[self.symtab[0][name]['Loc']] = \
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
nr(nt='LAMBDA', t=None)
del self.symtab[0][name]
if report:
raise EParseAlreadyDefined(self)
if self.tok[0] in ('=', ';'):
# This is a variable definition
if typ is None: # Typeless variables are not allowed
raise EParseSyntax(self)
if self.tok[0] == '=':
self.NextToken()
if self.extendedglobalexpr:
self.disallowglobalvars = True # Disallow forward globals.
# Mark backtracking position
pos = self.pos
errorpos = self.errorpos
tok = self.tok
try:
value = self.Parse_simple_expr()
self.expect(';')
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
value.Simple = True # Success - mark it as simple
except EParse:
# Backtrack
self.pos = pos
self.errorpos = errorpos
self.tok = tok
# Use advanced expression evaluation.
value = self.Parse_expression()
self.expect(';')
self.disallowglobalvars = False # Allow forward globals again.
else:
# Use LSL's dull global expression.
value = self.Parse_simple_expr()
self.expect(';')
value.Simple = True
else: # must be semicolon
value = None
assert self.scopeindex == 0
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
decl = nr(nt='DECL', t=typ, name=name, scope=0)
if value is not None:
value = self.autocastcheck(value, typ)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
decl.ch = [value]
self.NextToken()
self.AddSymbol('v', 0, name, Loc=len(self.tree), Type=typ)
self.tree.append(decl)
elif self.tok[0] == '(':
# This is a function definition
self.NextToken()
self.PushScope()
params = self.Parse_optional_param_list()
self.expect(')')
self.NextToken()
self.localevents = None
self.locallabels = set()
force_inline = False
if (self.enable_inline and self.tok[0] == 'IDENT'
and self.tok[1] == 'inline'):
self.NextToken()
force_inline = True
body = self.Parse_code_block(typ)
del self.locallabels
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
if typ and not getattr(body, 'LIR', False): # is LastIsReturn flag set?
self.errorpos = self.closebrace
raise EParseCodePathWithoutRet(self)
paramscope = self.scopeindex
self.AddSymbol('f', 0, name, Loc=len(self.tree), Type=typ,
Inline=force_inline,
ParamTypes=params[0], ParamNames=params[1])
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
self.tree.append(nr(nt='FNDEF', t=typ, name=name, scope=0,
pscope=paramscope, ptypes=params[0], pnames=params[1],
ch=[body]))
self.PopScope()
assert self.scopeindex == 0
else:
raise EParseSyntax(self)
pass
def Parse_states(self):
"""Grammar parsed here:
states: LAMBDA | states state
state: state_header '{' events '}'
state_header: DEFAULT | STATE IDENT
(but we enforce DEFAULT to be the first token found, meaning there will
be at least one state and the first must be DEFAULT as in the original
grammar)
"""
self.expect('DEFAULT')
while True:
if self.tok[0] != 'DEFAULT' and self.tok[0] != 'STATE':
return
if self.tok[0] == 'DEFAULT':
name = 'default'
else:
self.NextToken()
if self.tok[0] != 'IDENT':
raise EParseSyntax(self)
name = self.tok[1]
if name in self.symtab[self.scopeindex]:
raise EParseAlreadyDefined(self)
assert self.scopeindex == 0
self.AddSymbol('s', 0, name, Loc=len(self.tree))
self.NextToken()
self.expect('{')
self.NextToken()
self.localevents = set()
events = self.Parse_events()
del self.localevents
self.expect('}')
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
self.tree.append(nr(nt='STDEF', t=None, name=name, scope=0,
ch=events))
self.NextToken()
def Parse_script(self):
"""Parses the whole LSL script
Grammar parsed here:
script: globals states EOF
"""
# We need a table of undefined jump references, to check later,
# as jumps are local, not global, and allow forward definitions.
# This avoids making one more pass, or making the first pass more
# detailed unnecessarily.
self.jump_lookups = []
self.globalmode = True
self.Parse_globals()
self.globalmode = False
self.Parse_states()
self.expect('EOF')
assert len(self.scopestack) == 1 and self.scopestack[0] == 0
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
# Check the pending jump targets to assign them the scope of the label.
for tgt in self.jump_lookups:
self.scopestack = tgt[1]
self.scopeindex = self.scopestack[-1]
sym = self.FindSymbolPartial(tgt[0], MustBeLabel = True)
if sym is None:
self.errorpos = tgt[2]
raise EParseUndefined(self)
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
tgt[3].scope = sym['Scope']
sym['ref'] += 1
del self.jump_lookups # Finished with it.
self.scopestack = [0]
def Parse_single_expression(self):
"""Parse the script as an expression, Used by lslcalc.
Grammar parsed here:
script: expression EOF
"""
value = self.Parse_expression()
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
self.tree.append(nr(nt='EXPR', t=value.t, ch=[value]))
self.expect('EOF')
return
def BuildTempGlobalsTable(self):
"""Build an approximate globals table.
If the script syntax is correct, the globals table will be accurate.
If it is not, it may contain too many or too few symbols (normally the
latter). This globals table is not the normal globals in the symbol
table; it's just needed to resolve forward references. It's temporary.
The grammar is approximately:
script: globals states
globals: [global [global [...]]]
global: [TYPE] IDENT '(' TYPE anytoken [',' TYPE anytoken [...]]
anytoken_except_comma balanced_braces_or_anything_else
| TYPE IDENT [anytoken_except_semicolon [...]] ';'
states: state [state [...]]
state: (DEFAULT | STATE IDENT) balanced_braces_or_anything_else
"""
ret = self.funclibrary.copy() # The library functions go here too.
# If there's a syntax error, that's not our business. We just return
# what we have so far. Doing a proper parse will determine the exact
# location and cause.
# Here we don't even care if it's duplicate - that will be caught
# when adding to the real symbol table.
# Scan globals
try:
while self.tok[0] not in ('DEFAULT', 'EOF'):
typ = None
if self.tok[0] == 'TYPE':
typ = self.tok[1]
self.NextToken()
if self.tok[0] != 'IDENT':
return ret
name = self.tok[1]
self.NextToken()
if self.tok[0] == '(':
# Function call
self.NextToken()
params = []
if self.tok[0] != ')':
while True:
if self.tok[0] != 'TYPE':
return ret
params.append(self.tok[1])
self.NextToken()
self.NextToken() # not interested in parameter names
if self.tok[0] != ',':
break
self.NextToken()
self.NextToken()
if self.tok[0] == 'IDENT' and self.tok[1] == 'inline':
self.NextToken()
if self.tok[0] != '{':
return ret
self.NextToken() # Enter the first brace
bracelevel = 1
while bracelevel and self.tok[0] != 'EOF':
if self.tok[0] == '{':
bracelevel += 1
elif self.tok[0] == '}':
bracelevel -= 1
self.NextToken()
ret[name] = {'Kind':'f', 'Type':typ, 'ParamTypes':params,
'uns':True}
elif typ is None:
return ret # A variable needs a type
else:
# No location info but none is necessary for forward
# declarations.
ret[name] = {'Kind':'v','Type':typ,'Scope':0}
while self.tok[0] != ';': # Don't stop to analyze what's before the ending ';'
if self.tok[0] == 'EOF':
return ret
self.NextToken()
self.NextToken()
except EParseUEOF:
return ret
# Scan states
while True:
if self.tok[0] not in ('DEFAULT', 'STATE'):
return ret # includes EOF i.e. this is the normal return
if self.tok[0] == 'STATE':
self.NextToken()
if self.tok[0] != 'IDENT':
return ret
name = self.tok[1]
else:
name = 'default'
# No location info but none is necessary for forward declarations.
ret[name] = {'Kind':'s'}
self.NextToken()
if self.tok[0] != '{':
return ret
self.NextToken() # Enter the first brace
bracelevel = 1
while bracelevel and self.tok[0] != 'EOF':
if self.tok[0] == '{':
bracelevel += 1
elif self.tok[0] == '}':
bracelevel -= 1
self.NextToken()
def parse(self, script, options = (), filename = '<stdin>', lib = None):
"""Parse the given string with the given options.
If given, lib replaces the library passed in __init__.
filename is the filename of the current file, for error reporting.
'<stdin>' means errors in this file won't include a filename.
#line directives change the filename.
This function also builds the temporary globals table.
"""
if lib is None:
lib = self.lib
self.events = lib[0].copy()
self.constants = lib[1].copy()
self.funclibrary = lib[2].copy()
self.TypeToExtractionFunction.clear()
for name in self.funclibrary:
fn = self.funclibrary[name]
if 'ListTo' in fn:
self.TypeToExtractionFunction[fn['ListTo']] = name
self.filename = filename
2020-11-08 18:28:57 -07:00
script = any2str(script, 'utf8')
self.script = script
self.length = len(script)
self.keywords = self.base_keywords.copy()
self.labelcnt = 0
# Options
# Extended expressions in globals (needs support from the optimizer to work)
self.extendedglobalexpr = 'extendedglobalexpr' in options
# Extended typecast syntax (typecast as a regular unary operator)
self.extendedtypecast = 'extendedtypecast' in options
# Extended assignment operators: |= &= <<= >>=
self.extendedassignment = 'extendedassignment' in options
# Add explicit type casts when implicit (the output module takes care of
# the correctness of the output)
self.explicitcast = 'explicitcast' in options
# Allow string + key = string and key + string = string
self.allowkeyconcat = 'allowkeyconcat' in options
# Allow C style string composition of strings: "blah" "blah" = "blahblah"
self.allowmultistrings = 'allowmultistrings' in options
# Process preprocessor directives (especially #pragma and #line).
self.processpre = 'processpre' in options
2015-02-26 20:46:23 -07:00
# TODO: Allow pure C-style string escapes. This is low-priority.
#self.allowcescapes = 'allowcescapes' in options
# Enable switch statements.
self.enableswitch = 'enableswitch' in options
if self.enableswitch:
self.keywords |= self.switch_keywords
# Broken behaviour in the absence of a default: label in a switch stmt.
self.errmissingdefault = 'errmissingdefault' in options
# Allow brackets for assignment of list elements e.g. mylist[5]=4
self.lazylists = 'lazylists' in options
2015-03-03 16:40:07 -07:00
# This was once an idea, but it has been discarded because
# llListReplaceList requires the argument to be evaluated twice,
# so the function is unavoidable. Consider e.g. L[x++] = 3 expanded to
# L = llListReplaceList(L, [3], x++, x++).
# # Extend the list with integer zeros when lazylists is active and the
# # index is greater than the end of the list.
# self.lazylistcompat = 'lazylistcompat' in options
# Enable break/continue
self.breakcont = 'breakcont' in options
if self.breakcont:
self.keywords |= self.brkcont_keywords
# Stack to track the labels for break targets, their scope table index,
# and whether they are used.
# Elements are sublist with 0 = destination label name, 1 = scope for
# that label, and 2 = reference count of the label.
self.breakstack = []
# Stack to track the labels for continue targets, their scope index,
# and whether they are used.
self.continuestack = []
2014-12-13 05:12:02 -07:00
# Enable use of local labels with duplicate names
self.duplabels = 'duplabels' in options
# Shrink names. Activates duplabels automatically.
self.shrinknames = 'shrinknames' in options
# Allow a duplicate function definition to override the former,
# rather than reporting a duplicate identifier error.
self.funcoverride = 'funcoverride' in options
# This was an idea, but functions must return a type, and making the
# type suitable for the context is too much work.
# # Allow referencing undefined functions inside function definitions.
# self.allowundeffn = 'allowundeffn' in options
# Prettify a source file
self.prettify = 'prettify' in options
# We've decided to ditch support for optimization when the code
# includes a label as the immediate child of FOR, IF, DO or WHILE.
# If optimization is on, such a label will raise an error. That
# coding pattern is normally easy to work around anyway.
self.optenabled = 'optimize' in options
# Inline keyword
self.enable_inline = 'inline' in options
# Automated Processing friendly error messages
self.emap = 'emap' in options
# Symbol table:
# This is a list of all local and global symbol tables.
# The first element (0) is the global scope. Each symbol table is a
# dictionary of symbols, whose elements are in turn dictionaries of
# attributes. Each has a 'Kind', which can be:
# 'v' for variable, 'c' for constant, 'f' for function, 'l' for label,
# 's' for state, or 'e' for event. Some have a 'Loc' indicating the
# location (index) of the definition in the tree's root.
# Variables and constants have 'Scope' and 'Type' (a string).
# Global variables also have 'Loc'.
# Variables that are parameters also have 'Param'.
# Functions have 'Type' (return type, a string) and 'ParamTypes' (a
# list of strings).
# User-defined functions also have 'Loc' and 'ParamNames' (a list of
# strings).
# Labels only have 'Scope'.
# States only have 'Loc'.
# Events have 'ParamTypes' and 'ParamNames', just like UDFs.
# Other modules may add information if they need.
# Incorporate the library into the initial symbol table.
self.symtab = [self.funclibrary.copy()]
# Current scope index
self.scopeindex = 0
# Stack of scopes in which to look for a symbol as we parse
self.scopestack = [0]
if self.prettify:
# Add all constants to the blacklist
self.blacklist = list(u2str(i) for i in self.constants.keys())
# Remove TRUE and FALSE from keywords
self.keywords -= set(('TRUE', 'FALSE'))
# Some unit tests that reuse the parser object don't initialize the
# blacklist, so we don't rely on it being set.
if not hasattr(self, 'blacklist'):
self.blacklist = []
for name in self.blacklist:
# Add the blacklisted constants to the symbol table...
self.symtab[0][name] = {'Kind':'c', 'Scope':0, 'W':False,
'Type':lslcommon.PythonType2LSL[type(self.constants[name])]}
# ... and remove them from the list of substitutable constants.
del self.constants[name]
assert not self.prettify or len(self.constants) == 0
# Last preprocessor __FILE__. <stdin> means the current file.
self.lastFILE = '<stdin>'
# List of preprocessor #line directives.
self.linedir = []
# List of tuples (position, exception) where suspicious state change
# statements or returns with void expressions happen. These can only
# be detected when the 'else' is found.
self.PruneBug = []
# This is a small hack to prevent circular definitions in globals when
# extended expressions are enabled. When false (default), forward
# globals are allowed; if true, only already seen globals are permitted.
self.disallowglobalvars = False
2019-01-01 14:44:54 -07:00
# Hack to determine where to allow void expressions.
self.allowVoid = False
# Globals and labels can be referenced before they are defined. That
# includes states.
#
# Our first approach was going to be to build a list that keeps track of
# undefined references, to check them after parsing. But that has a big
# problem: expressions need to know the types of the arguments in order
# to give appropriate errors if they don't suit the operand, and in
# order to mark and check the types appropriately. But we don't know the
# types of the globals that haven't been found yet. Therefore, sticking
# to this approach would mean to scan the tree for every expression with
# a pending reference, fixing up every node upstream with the correct
# type with the possibility to find a type mismatch in a place for which
# we have no location info.
#
# For that reason, we change the strategy. We still don't want to do
# two full or almost full passes of the parser, nitpicking on every
# detail. But given LSL's structure, it's relatively easy to do a fast
# incomplete parsing pass, gathering globals with their types and
# function arguments. And that's what we do.
self.scanglobals = True # Tell the lexer not to process directives
self.pos = self.errorpos = 0
self.linestart = True
self.tok = self.GetToken()
self.globals = self.BuildTempGlobalsTable() if not lslcommon.IsCalc \
else self.funclibrary.copy()
# Restart
self.scanglobals = False
self.pos = self.errorpos = 0
self.linestart = True
self.tok = self.GetToken()
# Reserve spots at the beginning for functions we add
Change the AST node type from dict to object That was long overdue. Obviously, this is a large commit. The new nr (node record) class has built-in dump capabilities, rather than using print_node(). SEF always exists now, and is a boolean, rather than using the existence of SEF as the flag. This was changed for sanity. However, other flags like 'X' are still possibly absent, and in some cases the absence itself has meaning (in the case of 'X', its absence means that the node has not yet been analyzed). Similarly, an event is distinguished from a UDF by checking for the existence of the 'scope' attribute. This trick works because events are not in the symbol table therefore they have no scope. But this should probably be changed in future to something more rational and faster. A few minor bugfixes were applied while going through the code. - Some tabs used as Unicode were written as byte strings. Add the u'\t' prefix. - After simplifying a%1 -> a&0, fold again the node and return. It's not clear why it didn't return, and whether it depended on subsequent passes (e.g. after DCR) for possibly optimizing out the result. Now we're sure. - A few places lacked a SEF declaration. - Formatting changes to split lines that spilled the margin. - Some comment changes. - Expanded lazy_list_set definition while adapting it to object format. The plan was to re-compress it after done, but decided to leave it in expanded form. - Added a few TODOs & FIXMEs, resisting the temptation to fix them in the same commit: - TODO: ~-~-~-expr -> expr + -3. - FIXME: Now that we have CompareTrees, we can easily check if expr + -expr cancels out and remove a TODO. Low-hanging fruit. - TODO: Check what we can do when comparing non-SEF and non-CONST values in '>' (current code relies on converting '>' to '<' for applying more optimizations, but that may miss some opportunities). - FIXME: Could remove one comparison in nt == '&&' or nt == '||'. Low-hanging fruit.
2018-03-27 15:19:08 -07:00
self.tree = [nr(nt='LAMBDA', t=None)]
self.usedspots = 0
# Start the parsing proper
if lslcommon.IsCalc:
self.Parse_single_expression()
else:
self.Parse_script()
# No longer needed. The data is already in self.symtab[0].
del self.globals
del self.scopestack
if self.enable_inline:
from lslopt import lslinliner
lslinliner.inliner().inline(self.tree, self.symtab)
treesymtab = self.tree, self.symtab
del self.tree
del self.symtab
return treesymtab
def parsefile(self, filename, options = set(), lib = None):
"""Convenience function to parse a file rather than a string."""
f = open(filename, 'r')
try:
script = f.read()
finally:
f.close()
return self.parse(script, options, filename, lib)
def __init__(self, lib = None):
"""Initialization of library and lazy compilation.
lib is a tuple of three dictionaries: events, constants and functions,
in the format returned by lslloadlib.LoadLibrary().
"""
self.parse_directive_re = None
self.lib = lib if lib is not None else ({}, {}, {})