Files
LSL-PyOptimizer/lslopt/lsldeadcode.py

546 lines
22 KiB
Python
Raw Normal View History

# (C) Copyright 2015-2018 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/>.
# Dead Code Removal optimization
2014-08-10 02:07:00 +02:00
import lslfuncs
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.
2018-03-28 00:19:08 +02:00
from lslcommon import nr
2014-08-10 02:07:00 +02:00
class deadcode(object):
def MarkReferences(self, node):
"""Marks each node it passes through as executed (X), and each variable
as read (R) (with count) and/or written (W) (with node where it is, or
False if written more than once) as appropriate. Traces execution to
determine if any part of the code is never executed.
"""
# The 'X' key, when present, indicates whether a node is executed.
# Its value means whether this instruction will proceed to the next
# (True: it will; False: it won't).
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.
2018-03-28 00:19:08 +02:00
if hasattr(node, 'X'):
return node.X # branch already analyzed
2014-08-10 02:07:00 +02:00
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.
2018-03-28 00:19:08 +02:00
nt = node.nt
child = node.ch
2014-08-10 02:07:00 +02:00
# Control flow statements
if nt == 'STSW':
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.
2018-03-28 00:19:08 +02:00
node.X = False # Executed but path-breaking.
sym = self.symtab[0][node.name]
if not hasattr(self.tree[sym['Loc']], 'X'):
2014-08-10 02:07:00 +02:00
self.MarkReferences(self.tree[sym['Loc']])
return False
if nt == 'JUMP':
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.
2018-03-28 00:19:08 +02:00
node.X = False # Executed but path-breaking.
sym = self.symtab[node.scope][node.name]
2014-08-10 02:07:00 +02:00
if 'R' in sym:
sym['R'] += 1
else:
sym['R'] = 1
return False
if nt == 'RETURN':
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.
2018-03-28 00:19:08 +02:00
node.X = False # Executed but path-breaking.
2014-08-10 02:07:00 +02:00
if child:
self.MarkReferences(child[0])
return False
if nt == 'IF':
# "When you get to a fork in the road, take it."
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.
2018-03-28 00:19:08 +02:00
node.X = None # provisional value, refined later
2014-08-10 02:07:00 +02:00
self.MarkReferences(child[0])
condnode = child[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.
2018-03-28 00:19:08 +02:00
if condnode.nt == 'CONST':
if lslfuncs.cond(condnode.value):
2014-08-10 02:07:00 +02:00
# TRUE - 'then' branch always executed.
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.
2018-03-28 00:19:08 +02:00
node.X = self.MarkReferences(child[1])
return node.X
2014-08-10 02:07:00 +02:00
elif len(child) == 3:
# FALSE - 'else' branch always executed.
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.
2018-03-28 00:19:08 +02:00
node.X = self.MarkReferences(child[2])
return node.X
2014-08-10 02:07:00 +02:00
# else fall through
else:
cont = self.MarkReferences(child[1])
if len(child) == 3:
if not cont:
cont = self.MarkReferences(child[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.
2018-03-28 00:19:08 +02:00
node.X = cont
2014-08-10 02:07:00 +02:00
return cont
self.MarkReferences(child[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.
2018-03-28 00:19:08 +02:00
node.X = True
2014-08-10 02:07:00 +02:00
return True
if nt == 'WHILE':
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.
2018-03-28 00:19:08 +02:00
node.X = None # provisional value, refined later
2014-08-10 02:07:00 +02:00
self.MarkReferences(child[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.
2018-03-28 00:19:08 +02:00
if child[0].nt == 'CONST':
if lslfuncs.cond(child[0].value):
2014-08-10 02:07:00 +02:00
# Infinite loop - unless it returns, it stops
# execution. But it is executed itself.
self.MarkReferences(child[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.
2018-03-28 00:19:08 +02:00
node.X = False
return node.X
2014-08-10 02:07:00 +02:00
# else the inside isn't executed at all, so don't mark it
else:
self.MarkReferences(child[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.
2018-03-28 00:19:08 +02:00
node.X = True
2014-08-10 02:07:00 +02:00
return True
if nt == 'DO':
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.
2018-03-28 00:19:08 +02:00
node.X = None # provisional value, refined later
2014-08-10 02:07:00 +02:00
if not self.MarkReferences(child[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.
2018-03-28 00:19:08 +02:00
node.X = False
2014-08-10 02:07:00 +02:00
return False
self.MarkReferences(child[1])
# It proceeds to the next statement unless it's an infinite loop
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.
2018-03-28 00:19:08 +02:00
node.X = not (child[1].nt == 'CONST' and lslfuncs.cond(child[1].value))
return node.X
2014-08-10 02:07:00 +02:00
if nt == 'FOR':
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.
2018-03-28 00:19:08 +02:00
node.X = None # provisional value, refined later
2014-08-10 02:07:00 +02:00
self.MarkReferences(child[0])
self.MarkReferences(child[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.
2018-03-28 00:19:08 +02:00
if child[1].nt == 'CONST':
if lslfuncs.cond(child[1].value):
2014-08-10 02:07:00 +02:00
# Infinite loop - unless it returns, it stops
# execution. But it is executed itself.
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.
2018-03-28 00:19:08 +02:00
node.X = False
2014-08-10 02:07:00 +02:00
self.MarkReferences(child[3])
self.MarkReferences(child[2]) # this can't stop execution
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.
2018-03-28 00:19:08 +02:00
return node.X
2014-08-10 02:07:00 +02:00
# else the body and the iterator aren't executed at all, so
# don't mark them
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.
2018-03-28 00:19:08 +02:00
node.X = True
2014-08-10 02:07:00 +02:00
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.
2018-03-28 00:19:08 +02:00
node.X = True
2014-08-10 02:07:00 +02:00
self.MarkReferences(child[3])
self.MarkReferences(child[2])
# Mark the EXPRLIST as always executed, but not the subexpressions.
# That forces the EXPRLIST (which is a syntactic requirement) to be
# kept, while still simplifying the contents properly.
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.
2018-03-28 00:19:08 +02:00
child[2].X = True
2014-08-10 02:07:00 +02:00
return True
if nt == '{}':
# Go through each statement in turn. If one stops execution,
# continue reading until either we find a used label (and resume
# execution) or reach the end (and return False). Otherwise return
# True.
continues = 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.
2018-03-28 00:19:08 +02:00
node.X = None # provisional
2014-08-10 02:07:00 +02:00
for stmt in child:
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.
2018-03-28 00:19:08 +02:00
if continues or stmt.nt == '@':
2014-08-10 02:07:00 +02:00
continues = self.MarkReferences(stmt)
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.
2018-03-28 00:19:08 +02:00
node.X = continues
2014-08-10 02:07:00 +02:00
return continues
if nt == 'FNCALL':
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.
2018-03-28 00:19:08 +02:00
node.X = None # provisional
sym = self.symtab[0][node.name]
2014-08-10 02:07:00 +02:00
fdef = self.tree[sym['Loc']] if 'Loc' in sym else None
for idx in xrange(len(child)-1, -1, -1):
# Each element is a "write" on the callee's parameter.
# E.g. f(integer a, integer b) { f(2,3); } means 2, 3 are
# writes to a and b.
self.MarkReferences(child[idx])
if fdef 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.
2018-03-28 00:19:08 +02:00
psym = self.symtab[fdef.pscope][fdef.pnames[idx]]
2014-08-10 02:07:00 +02:00
if 'W' in psym:
psym['W'] = False
else:
psym['W'] = child[idx]
if 'Loc' in sym:
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.
2018-03-28 00:19:08 +02:00
if not hasattr(self.tree[sym['Loc']], 'X'):
2014-08-10 02:07:00 +02:00
self.MarkReferences(self.tree[sym['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.
2018-03-28 00:19:08 +02:00
node.X = self.tree[sym['Loc']].X
2014-08-10 02:07:00 +02:00
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.
2018-03-28 00:19:08 +02:00
node.X = 'stop' not in sym
2014-08-10 02:07:00 +02:00
# Note that JUMP analysis is incomplete. To do it correctly, we
# should follow the jump right to its destination, in order to know
# if that branch leads to a RETURN or completely stops the event.
# With our code structure, following the JUMP is unfeasible.
# For that reason, we can't track whether a branch ends in RETURN
# or in something more powerful like a script reset, in order to
# propagate it through the function definition to the caller.
#
# In practice, this means that the caller can't distinguish this:
# fn() { return; }
# from this:
# fn() { llResetScript(); }
# and therefore, invocations of the function that are followed by
# code can't know whether that code is dead or not.
#
# What does that have to do with jumps? Well, imagine this:
# fn(integer x) { if (x) jump x1; else jump x2;
# @x1; return; @x2; llResetScript(); }
# What of the branches of if() is taken, depends on where the jumps
# lead to. Assuming the last one always is wrong, because it would
# mark in the caller code that may execute, as dead, e.g. here:
# fn2() { fn(); x = 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.
2018-03-28 00:19:08 +02:00
return node.X
2014-08-10 02:07:00 +02:00
if nt == 'DECL':
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.
2018-03-28 00:19:08 +02:00
sym = self.symtab[node.scope][node.name]
2014-08-10 02:07:00 +02:00
if child is not None:
sym['W'] = child[0]
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.
2018-03-28 00:19:08 +02:00
sym['W'] = nr(nt='CONST', t=node.t,
value=self.DefaultValues[node.t])
2014-08-10 02:07:00 +02:00
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.
2018-03-28 00:19:08 +02:00
node.X = True
2014-08-10 02:07:00 +02:00
if child 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.
2018-03-28 00:19:08 +02:00
if hasattr(child[0], 'orig'):
orig = child[0].orig
self.MarkReferences(orig)
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.
2018-03-28 00:19:08 +02:00
child[0].X = orig.X
if orig.nt == 'LIST':
# Add fake writes to variables used in list elements in
# 'orig', so they don't get deleted (Issue #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.
2018-03-28 00:19:08 +02:00
for subnode in orig.ch:
if subnode.nt == 'IDENT':
# can only happen 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.
2018-03-28 00:19:08 +02:00
assert subnode.scope == 0
sym = self.symtab[0][subnode.name]
sym['W'] = 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.
2018-03-28 00:19:08 +02:00
self.tree[sym['Loc']].X = True
elif subnode.nt in ('VECTOR', 'ROTATION'):
for sub2node in subnode.ch:
if sub2node.nt == 'IDENT':
2018-03-23 22:05:05 +01:00
# can only happen 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.
2018-03-28 00:19:08 +02:00
assert sub2node.scope == 0
sym = self.symtab[0][sub2node.name]
sym['W'] = 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.
2018-03-28 00:19:08 +02:00
self.tree[sym['Loc']].X = True
2014-08-10 02:07:00 +02:00
else:
self.MarkReferences(child[0])
return True
# ---- Starting here, all node types return through the bottom
# (except '=').
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.
2018-03-28 00:19:08 +02:00
node.X = None # provisional
2014-08-10 02:07:00 +02:00
if nt in self.assign_ops or nt in ('--V', '++V', 'V++', 'V--'):
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.
2018-03-28 00:19:08 +02:00
ident = node.ch[0]
if ident.nt == 'FLD':
ident = ident.ch[0]
assert ident.nt == 'IDENT'
sym = self.symtab[ident.scope][ident.name]
if ident.scope == 0:
2014-08-10 02:07:00 +02:00
# Mark the global first.
self.MarkReferences(self.tree[sym['Loc']])
# In any case, this is at least the second write, so mark it as such
# (SSA would be a plus for this to be optimal)
sym['W'] = False
if nt == '=':
# Prevent the first node from being mistaken as a read, by
# recursing only on the RHS node.
self.MarkReferences(child[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.
2018-03-28 00:19:08 +02:00
node.X = True
2014-08-10 02:07:00 +02:00
return True
elif nt == 'FLD':
# Mark this variable as referenced by a Field (recursing will mark
# the ident as read later)
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.
2018-03-28 00:19:08 +02:00
self.symtab[child[0].scope][child[0].name]['Fld'] = True
2014-08-10 02:07:00 +02:00
elif nt == 'IDENT':
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.
2018-03-28 00:19:08 +02:00
sym = self.symtab[node.scope][node.name]
2014-08-10 02:07:00 +02:00
# Mark global if it's 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.
2018-03-28 00:19:08 +02:00
if 'W' not in sym and node.scope == 0:
2014-08-10 02:07:00 +02:00
self.MarkReferences(self.tree[sym['Loc']])
# Increase read counter
if 'R' in sym:
sym['R'] += 1
else:
sym['R'] = 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.
2018-03-28 00:19:08 +02:00
node.X = True
2014-08-10 02:07:00 +02:00
if child is not None:
for subnode in child:
self.MarkReferences(subnode)
return True
def SymbolReplacedOrDeleted(self, curnode):
"""If the given node's name must be simplified, that is, replaced or
deleted (deleted if declaration, replaced if identifier), it returns
the symbol table entry. Otherwise it returns False.
"""
# 'W':False means written more than once, i.e. not only in the
# declaration. Variables written more than once can't be
# simplified by this code.
# If not written more than once:
# For expressions:
# - Remove expressions read only once, replacing the value, but only
# if the readers are not fields of vectors or rotations, and if
# they are side-effect free.
# For constants:
# - Remove lists, vectors and rotations read only once.
# - Floats are removed if their value has no decimals or if
# used no more than N times (for some N).
# - Strings, keys and integers are just removed.
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.
2018-03-28 00:19:08 +02:00
sym = self.symtab[curnode.scope][curnode.name]
2014-08-10 02:07:00 +02:00
if 'R' not in sym:
return True # if not used, it can be removed
# Event parameters do not have 'W' in sym.
if 'W' not in sym:
return False
if sym['W'] is not False:
node = sym['W']
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.
2018-03-28 00:19:08 +02:00
nt = node.nt
2014-08-10 02:07:00 +02:00
if nt == 'CONST':
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.
2018-03-28 00:19:08 +02:00
tcurnode = curnode.t
2014-08-10 02:07:00 +02:00
if tcurnode in ('integer', 'string', 'key'):
return sym
if tcurnode == '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.
2018-03-28 00:19:08 +02:00
if sym['R'] <= 3 or type(node.value) == int:
2014-08-10 02:07:00 +02:00
return sym
elif tcurnode == '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.
2018-03-28 00:19:08 +02:00
or tcurnode == 'list' and len(node.value) <= 3:
2014-08-10 02:07:00 +02:00
if sym['R'] <= 1:
return sym
elif tcurnode == '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.
2018-03-28 00:19:08 +02:00
or tcurnode == 'list' and len(node.value) <= 4:
2014-08-10 02:07:00 +02:00
if sym['R'] <= 1:
return sym
return False
# To replace expressions, they MUST be side-effect free, or they
# will be executed at a different time.
# Also, we can't safely replace expressions unless shrinknames is
# active. shrinknames assigns a different identifier to each
# variable, which avoids conflicts. Consider this scenario:
# integer i=2;
# default{state_entry(){integer j=i+1; integer i=4; llSleep(i+j);}}
# Replacing j with i+1 in llOwnerSay will produce wrong code because
# the name i is redefined after j is assigned. shrinknames prevents
# that.
# FIXME: EMERGENCY FIX: shrinknames is not enough guarantee. See nposerlv.lsl.
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.
2018-03-28 00:19:08 +02:00
#if not self.shrinknames or not node.SEF:
if True or not node.SEF:
2014-08-10 02:07:00 +02:00
return False
if nt not in ('VECTOR', 'ROTATION'):
# If it's an expression and the reference is to a field, we
# can't simplify. Consider e.g.:
# vector v = llGetVel(); llOwnerSay((string)v.z);
# However, if it's a field coming from a Vector or Rotation
# expression, we can embed the corresponding component, e.g.
# vector v = <i+1, i+2, i+3>; llOwnerSay((string)v.y);
# can be replaced with: llOwnerSay((string)((float)(i+2)));
if 'Fld' in sym:
return False
if sym['R'] == 1:
return sym
return False
def CleanNode(self, curnode):
"""Recursively checks if the children are used, deleting those that are
not.
"""
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.
2018-03-28 00:19:08 +02:00
if curnode.ch is None or (curnode.nt == 'DECL'
and curnode.scope == 0):
2014-08-10 02:07:00 +02:00
return
# NOTE: Should not depend on 'Loc', since the nodes that are the
# destination of 'Loc' are renumbered as we delete stuff from 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.
2018-03-28 00:19:08 +02:00
index = int(curnode.nt in self.assign_ops) # don't recurse into a lvalue
2014-08-10 02:07:00 +02:00
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.
2018-03-28 00:19:08 +02:00
while index < len(curnode.ch):
node = curnode.ch[index]
2014-08-10 02:07:00 +02:00
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.
2018-03-28 00:19:08 +02:00
if not hasattr(node, 'X'):
del curnode.ch[index]
2014-08-10 02:07:00 +02:00
continue
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.
2018-03-28 00:19:08 +02:00
nt = node.nt
2014-08-10 02:07:00 +02:00
if nt == 'DECL':
if self.SymbolReplacedOrDeleted(node):
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.
2018-03-28 00:19:08 +02:00
if not node.ch or node.ch[0].SEF:
del curnode.ch[index]
2014-08-10 02:07:00 +02:00
continue
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.
2018-03-28 00:19:08 +02:00
node = curnode.ch[index] = nr(nt='EXPR', t=node.t,
ch=[self.Cast(node.ch[0], node.t)])
2014-08-10 02:07:00 +02:00
elif nt == 'FLD':
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.
2018-03-28 00:19:08 +02:00
sym = self.SymbolReplacedOrDeleted(node.ch[0])
2014-08-10 02:07:00 +02:00
if sym:
value = sym['W']
# Mark as executed, so it isn't optimized out.
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.
2018-03-28 00:19:08 +02:00
value.X = True
fieldidx = 'xyzs'.index(node.fld)
if value.nt == 'CONST':
value = value.value[fieldidx]
value = nr(nt='CONST', X=True, SEF=True,
t=self.PythonType2LSL[type(value)], value=value)
2014-08-10 02:07:00 +02:00
value = self.Cast(value, 'float')
SEF = True
else: # assumed VECTOR or ROTATION per SymbolReplacedOrDeleted
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.
2018-03-28 00:19:08 +02:00
SEF = value.SEF
value = self.Cast(value.ch[fieldidx], 'float')
# Replace it
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.
2018-03-28 00:19:08 +02:00
node = curnode.ch[index] = value
node.SEF = SEF
2014-08-10 02:07:00 +02:00
elif nt == 'IDENT':
sym = self.SymbolReplacedOrDeleted(node)
if sym:
# Mark as executed, so it isn't optimized out.
# Make shallow copy.
# TODO: Needs more analysis to see if it's correct or not.
# (See constant_anomaly.lsl)
new = sym['W'].copy()
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.
2018-03-28 00:19:08 +02:00
if hasattr(new, 'orig'):
del new.orig
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.
2018-03-28 00:19:08 +02:00
new.X = True
2014-08-10 02:07:00 +02:00
# this part makes no sense?
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.
2018-03-28 00:19:08 +02:00
#new.SEF = sym['W'].SEF
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.
2018-03-28 00:19:08 +02:00
if new.t != node.t:
new = self.Cast(new, node.t)
curnode.ch[index] = node = new
2014-08-10 02:07:00 +02:00
# Delete orig if present, as we've eliminated the original
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.
2018-03-28 00:19:08 +02:00
#if hasattr(sym['W'], 'orig'):
# del sym['W'].orig
2014-08-10 02:07:00 +02:00
elif nt in self.assign_ops:
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.
2018-03-28 00:19:08 +02:00
ident = node.ch[0]
if ident.nt == 'FLD':
ident = ident.ch[0]
2014-08-10 02:07:00 +02:00
sym = self.SymbolReplacedOrDeleted(ident)
if sym:
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.
2018-03-28 00:19:08 +02:00
node = curnode.ch[index] = self.Cast(node.ch[1], node.t)
2014-08-10 02:07:00 +02:00
elif nt in ('IF', 'WHILE', 'DO', 'FOR'):
# If the mandatory statement is to be removed, replace it
# with a ; to prevent leaving the statement empty.
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.
2018-03-28 00:19:08 +02:00
child = node.ch
2014-08-10 02:07:00 +02:00
idx = 3 if nt == 'FOR' else 0 if nt == 'DO' else 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.
2018-03-28 00:19:08 +02:00
if not hasattr(child[idx], 'X'):
child[idx] = nr(nt=';', t=None, X=True, SEF=True)
if nt == 'DO' and not hasattr(child[1],'X'):
2014-08-10 02:07:00 +02:00
# Mandatory condition but not executed - replace
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.
2018-03-28 00:19:08 +02:00
child[1] = nr(nt='CONST', X=True, SEF=True, t='integer',
value=0)
2014-08-10 02:07:00 +02:00
self.CleanNode(node)
index += 1
def RemoveDeadCode(self):
"""Simple reference-based dead code removal. It also performs a
simplified form of constant propagation, taking advantage of the fact
that it analyzes the code flow.
"""
# TODO: Converting to SSA first would facilitate DCR.
# The SSA should be followed by constant and expression propagation,
# then constant folding and then dead code removal.
# Start at state default and mark everything referenced from there.
# At the end, unreferenced globals and states will be removed.
# We assume all events in a state are executed.
#
# This may not be the case, e.g. a target()/no_target() without
# llTarget, sensor()/no_sensor() without llSensor(), listen() without
# llListen, timer without llSetTimerEvent, etc. but we're not that
# sophisticated (yet).
# TODO: Inlining of functions that are a single 'return' line.
# TODO: Remove empty events if they are side effect-free.
# Note that e.g. touch/* events are not SEF, because their presence
# causes the hand to be shown, so we need a list of SEF events.
if lslfuncs.lslcommon.IsCalc:
# Do nothing if in calculator mode (there's no default event
# and it crashes without this)
return
2014-08-10 02:07:00 +02:00
statedef = self.tree[self.symtab[0]['default']['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.
2018-03-28 00:19:08 +02:00
assert statedef.nt == 'STDEF' and statedef.name == 'default'
2014-08-10 02:07:00 +02:00
self.MarkReferences(statedef)
# Track removal of global lines, to reasign locations later.
LocMap = range(len(self.tree))
GlobalDeletions = []
2014-08-10 02:07:00 +02:00
# Perform the removal
idx = 0
while idx < len(self.tree):
# Globals are special.
# We need to track the locations too.
node = self.tree[idx]
delete = 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.
2018-03-28 00:19:08 +02:00
if not hasattr(node, 'X'):
2014-08-10 02:07:00 +02:00
delete = 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.
2018-03-28 00:19:08 +02:00
elif node.nt == 'DECL':
2014-08-10 02:07:00 +02:00
delete = self.SymbolReplacedOrDeleted(node)
if delete:
# Mark the symbol for later deletion from symbol table.
# We can't remove it here because there may be more references
# that we will remove in CleanNode later, that hold the
# original 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.
2018-03-28 00:19:08 +02:00
if node.nt == 'DECL' or node.nt == 'STDEF':
GlobalDeletions.append(node.name)
2014-08-10 02:07:00 +02:00
del self.tree[idx]
del LocMap[idx]
else:
idx += 1
self.CleanNode(node)
# Remove the globals now.
for name in GlobalDeletions:
del self.symtab[0][name]
del GlobalDeletions
2014-08-10 02:07:00 +02:00
# Reassign locations
for name in self.symtab[0]:
if name != -1:
sym = self.symtab[0][name]
if 'Loc' in sym:
try:
sym['Loc'] = LocMap.index(sym['Loc'])
except ValueError:
# Subtree deleted - delete the Loc
del sym['Loc']