This change implements a new bytecode compiler, based on a
transformation of the parse tree to an abstract syntax defined in
Parser/Python.asdl.
The compiler implementation is not complete, but it is in stable
enough shape to run the entire test suite excepting two disabled
tests.
* Use simpler, faster two pass algorithm for markblocks().
* Free the blocks variable if not NULL and exiting without change.
* Verify that the rest of the compiler has not set an exception.
* Make the test for tuple of constants less restrictive.
* Embellish the comment for chained conditional jumps.
Peepholer could be fooled into misidentifying a tuple_of_constants.
Added code to count consecutive occurrences of LOAD_CONST.
Use the count to weed out the misidentified cases.
Added a unittest.
Allows the lineno fixup code to remain simple and not have to deal with
multibyte codings.
* Add an assertion to that effect.
* Remove the XXX comment on the subject.
* Perform the code length check earlier.
* Eliminate the extra PyMem_Free() upon hitting an EXTENDED_ARG.
* Assert that the NOP count used in jump retargeting matches the NOPs
eliminated in the final step.
* Add an XXX note to indicate that more work is being to done to
handle linenotab with intervals > 255.
* Make a pass to eliminate NOPs. Produce code that is more readable,
more compact, and a tiny bit faster. Makes the peepholer more flexible
in the scope of allowable transformations.
* With Guido's okay, bumped up the magic number so that this patch gets
widely exercised before the alpha goes out.
files and not re-optimized upon import. Saves a bit of startup time while
still remaining decoupled from the rest of the compiler.
As a side benefit, handcoded bytecode is not run through the optimizer
when new code objects are created. Hopefully, a handcoder has already
created exactly what they want to have run.
(Idea suggested by Armin Rigo and Michael Hudson. Initially avoided
because of worries about compiler coupling; however, only the nexus
point needed to be moved so there won't be a conflict when the AST
branch is loaded.)
[ 1009560 ] Fix @decorator evaluation order
From the description:
Changes in this patch:
- Change Grammar/Grammar to require
newlines between adjacent decorators.
- Fix order of evaluation of decorators
in the C (compile.c) and python
(Lib/compiler/pycodegen.py) compilers
- Add better order of evaluation check
to test_decorators.py (test_eval_order)
- Update the decorator documentation in
the reference manual (improve description
of evaluation order and update syntax
description)
and the comment:
Used Brett's evaluation order (see
http://mail.python.org/pipermail/python-dev/2004-August/047835.html)
(I'm checking this in for Anthony who was having problems getting SF to
talk to him)
[ 1005248 ] new.code() not cleanly checking its arguments
using the result of new.code() can still destroy the sun, but merely
calling the function shouldn't any more.
I also rewrote the existing tests of new.code() to use vastly less
un-bogus arguments, and added tests for the previous insane behaviours.
The preceding case statement was missing a terminating "break" stmt,
so fell into the new code by mistake. This caused uncaught out-of-bounds
accesses to the "names" tuple, leading to a variety of insane behaviors.
(Code contributed by Jiwon Seo.)
The documentation portion of the patch is being re-worked and will be
checked-in soon. Likewise, PEP 289 will be updated to reflect Guido's
rationale for the design decisions on binding behavior (as described in
in his patch comments and in discussions on python-dev).
The test file, test_genexps.py, is written in doctest format and is
meant to exercise all aspects of the the patch. Further additions are
welcome from everyone. Please stress test this new feature as much as
possible before the alpha release.
and left shifts. (Thanks to Kalle Svensson for SF patch 849227.)
This addresses most of the remaining semantic changes promised by
PEP 237, except for repr() of a long, which still shows the trailing
'L'. The PEP appears to promise warnings for operations that
changed semantics compared to Python 2.3, but this is not
implemented; we've suffered through enough warnings related to
hex/oct literals and I think it's best to be silent now.
Fixes for three related bugs, including errors that caused a script to
be ignored without printing an error message. The key problem was a bad
interaction between syntax warnings and syntax errors. If an
exception was already set when a warning was issued, the warning could
clobber the exception.
The PyErr_Occurred() check in issue_warning() isn't entirely
satisfying (the caller should know whether there was already an
error), but a better solution isn't immediately obvious.
Bug fix candidate.
It depended on the previously removed basic block checker to
prevent a jump into the middle of the transformed block.
Clears SF 757818: tuple assignment -- SystemError: unknown opcode
The compiler was reseting the list comprehension tmpname counter for each function, but the symtable was using the same counter for the entire module. Repair by move tmpname into the symtable entry.
Bugfix candidate.
[ 708901 ] Lineno calculation sometimes broken
A one line patch to compile.c and a rather-more-than-one-line patch
to test_dis. Hey ho.
Possibly a backport candidate -- tho' lnotab is less used in 2.2...
* Can now test for basic blocks.
* Optimize inverted comparisions.
* Optimize unary_not followed by a conditional jump.
* Added a new opcode, NOP, to keep code size constant.
* Applied NOP to previous transformations where appropriate.
Note, the NOP would not be necessary if other functions were
added to re-target jump addresses and update the co_lnotab mapping.
That would yield slightly faster and cleaner bytecode at the
expense of optimizer simplicity and of keeping it decoupled
from the line-numbering structure.
Adds a single function to improve generated bytecode. Has a single line
attachment point, so it is completely de-coupled from both the compiler
and ceval.c.
Makes three simple transforms that do not require a basic block analysis
or re-ordering of code. Gives improved timings on pystone, pybench,
and any code using either "while 1" or "x,y=y,x".
"Unsigned" (i.e., positive-looking, but really negative) hex/oct
constants with a leading minus sign are once again properly negated.
The micro-optimization for negated numeric constants did the wrong
thing for such hex/oct constants. The patch avoids the optimization
for all hex/oct constants.
This needs to be backported to Python 2.2!
-DCALL_PROFILE: Count the number of function calls executed.
When this symbol is defined, the ceval mainloop and helper functions
count the number of function calls made. It keeps detailed statistics
about what kind of object was called and whether the call hit any of
the special fast paths in the code.
Optimization:
When we take the fast_function() path, which seems to be taken for
most function calls, and there is minimal frame setup to do, avoid
call PyEval_EvalCodeEx(). The eval code ex function does a lot of
work to handle keywords args and star args, free variables,
generators, etc. The inlined version simply allocates the frame and
copies the arguments values into the frame.
The optimization gets a little help from compile.c which adds a
CO_NOFREE flag to code objects that don't have free variables or cell
variables. This change allows fast_function() to get into the fast
path with fewer tests.
I measure a couple of percent speedup in pystone with this change, but
there's surely more that can be done.
A variety of changes from Michael Hudson to get the compiler working
with 2.3. The primary change is the handling of SET_LINENO:
# The set_lineno() function and the explicit emit() calls for
# SET_LINENO below are only used to generate the line number table.
# As of Python 2.3, the interpreter does not have a SET_LINENO
# instruction. pyassem treats SET_LINENO opcodes as a special case.
A few other small changes:
- Remove unused code from pycodegen and pyassem.
- Fix error handling in parsermodule. When PyParser_SimplerParseString()
fails, it sets an exception with detailed info. The parsermodule
was clobbering that exception and replacing it was a generic
"could not parse string" exception. Keep the original exception.
[#448679] Left to right
* Python/compile.c
(com_dictmaker): Reordered evaluation of dictionaries to follow strict
LTR evaluation.
* Lib/compiler/pycodegen.py
(CodeGenerator.visitDict): Reordered evaluation of dictionaries to
follow strict LTR evaluation.
* Doc/ref/ref5.tex
Documented the general LTR evaluation order idea.
* Misc/NEWS
Documented change in evaluation order of dictionaries.
Use a slightly different strategy to determine when not to call the line
trace function. This removes the need for the RETURN_NONE opcode, so
that's gone again. Update docs and comments to match.
Thanks to Neal and Armin!
Also add a test suite. This should have come with the original patch...
warning for 'global None', but that's either accompanied by an
assignment to None, which will trigger a warning, or not, in which
case it's harmless. :-)
[ 587993 ] SET_LINENO killer
Remove SET_LINENO. Tracing is now supported by inspecting co_lnotab.
Many sundry changes to document and adapt to this change.
currently return inconsistent results for ints and longs; in
particular: hex/oct/%u/%o/%x/%X of negative short ints, and x<<n that
either loses bits or changes sign. (No warnings for repr() of a long,
though that will also change to lose the trailing 'L' eventually.)
This introduces some warnings in the test suite; I'll take care of
those later.
Change the parser and compiler to use PyMalloc.
Only the files implementing processes that will request memory
allocations small enough for PyMalloc to be a win have been
changed, which are:-
- Python/compile.c
- Parser/acceler.c
- Parser/node.c
- Parser/parsetok.c
This augments the aggressive overallocation strategy implemented by
Tim Peters in PyNode_AddChild() [Parser/node.c], in reducing the
impact of platform malloc()/realloc()/free() corner case behaviour.
Such corner cases are known to be triggered by test_longexp and
test_import.
Jeremy Hylton, in accepting this patch, recommended this as a
bugfix candidate for 2.2. While the changes to Python/compile.c
and Parser/node.c backport easily (and could go in), the changes
to Parser/acceler.c and Parser/parsetok.c require other not
insignificant changes as a result of the differences in the memory
APIs between 2.3 and 2.2, which I'm not in a position to work
through at the moment. This is a pity, as the Parser/parsetok.c
changes are the most important after the Parser/node.c changes, due
to the size of the memory requests involved and their frequency.
is slow things down unnecessarily and make tracing much more verbose.
Something like
def f(n):
return [i for i in range(n) if i%2]
should have at most two SET_LINENO instructions, not four. When tracing,
the current line number should be printed once, not 2*n+1 times.
These built-in functions are replaced by their (now callable) type:
slice()
buffer()
and these types can also be called (but have no built-in named
function named after them)
classobj (type name used to be "class")
code
function
instance
instancemethod (type name used to be "instance method")
The module "new" has been replaced with a small backward compatibility
placeholder in Python.
A large portion of the patch simply removes the new module from
various platform-specific build recipes. The following binary Mac
project files still have references to it:
Mac/Build/PythonCore.mcp
Mac/Build/PythonStandSmall.mcp
Mac/Build/PythonStandalone.mcp
[I've tweaked the code layout and the doc strings here and there, and
added a comment to types.py about StringTypes vs. basestring. --Guido]
The old syntax suggested that a trailing comma was OK inside backticks,
but in fact (due to ideosyncrasies of pgen) it was not. Fix the grammar
to avoid the ambiguity. Fred: you may want to update the refman.
+ Redirect PyMem_{Del, DEL} to the object allocator's free() when
pymalloc is enabled. Needed so old extensions can continue to
mix PyObject_New with PyMem_DEL.
+ This implies that pgen needs to be able to see the PyObject_XYZ
declarations too. pgenheaders.h now includes Python.h. An
implication is that I expect obmalloc.o needs to get linked into
pgen on non-Windows boxes.
+ When PYMALLOC_DEBUG is defined, *all* Py memory API functions
now funnel through the debug allocator wrapper around pymalloc.
This is the default in a debug build.
+ That caused compile.c to fail: it indirectly mixed PyMem_Malloc
with raw platform free() in one place. This is verbotten.
type.__module__ problems (again?)
This simply initializes the __module__ local in a class statement from
the __name__ global. I'm not 100% sure that this is the correct fix,
although it usually does the right thing. The problem is that if the
class statement executes in a custom namespace, the __name__ global
may be taken from __builtins__, in which case it would have the value
__builtin__, or it may not exist at all (if the custom namespace also
has a custom __builtins__), in which case the class statement will
fail.
Nevertheless, unless someone finds a better solution, this is a 2.2.1
bugfix too.
Based on the patch from Danny Yoo. The fix is in exec_statement() in
ceval.c.
There are also changes to introduce use of PyCode_GetNumFree() in
several places.
The error for assignment to __debug__ used ste->ste_opt_lineno instead
of n->n_lineno. The latter was at best incorrect; often the slot was
uninitialized. Two fixes here: Use the correct lineno for the error.
Initialize ste_opt_lineno in PySymtable_New(); while there are no
current cases where it is referenced unless it has already been
assigned to, there is no harm in initializing it.
com_arglist(), symtable_check_unoptimized(), symtable_params(),
symtable_global(), symtable_list_comprehension():
Conversion of sprintf() to PyOS_snprintf() for buffer overrun
avoidance.
The symbol table pass didn't have an explicit case for the list_iter
node which is used only for a nested list comprehension. As a result,
the target of the list comprehension was treated as a use instead of
an assignment. Fix is to add a case to symtable_node() to handle
list_iter.
Also, rework and document a couple of the subtler implementation
issues in the symbol table pass. The symtable_node() switch statement
depends on falling through the last several cases, in order to handle
some of the more complicated nodes like atom. Add a comment
explaining the behavior before the first fall through case. Add a
comment /* fall through */ at the end of case so that it is explicitly
marked as such.
Move the for_stmt case out of the fall through logic, which simplifies
both for_stmt and default. (The default used the local variable start
to skip the first three nodes of a for_stmt when it fell through.)
Rename the flag argument to symtable_assign() to def_flag and add a
comment explaining its use:
The third argument to symatble_assign() is a flag to be passed to
symtable_add_def() if it is eventually called. The flag is useful
to specify the particular type of assignment that should be
recorded, e.g. an assignment caused by import.
Also minor tweaks to internal routines.
Use PyCF_MASK instead of explicit list of flags.
For the MAKE_CLOSURE opcode, the number of items popped off the stack
depends on both the oparg and the number of free variables for the
code object. Fix the code so it accounts for the free variables.
In com_classdef(), record an extra pop to account for the STORE call
after the BUILD_CLASS.
Get rid of some commented out debugging code in com_push() and
com_pop().
Factor string resize logic into helper routine com_check_size().
In com_addbyte(), remove redudant if statement after assert. (They
test the same condition.)
In several routines, use string macros instead of string functions.
"for <var> in <testlist> may no longer be a single test followed by
a comma. This solves SF bug #431886. Note that if the testlist
contains more than one test, a trailing comma is still allowed, for
maximum backward compatibility; but this example is not:
[(x, y) for x in range(10), for y in range(10)]
^
The fix involved creating a new nonterminal 'testlist_safe' whose
definition doesn't allow the trailing comma if there's only one test:
testlist_safe: test [(',' test)+ [',']]
compatibility, this required all places where an array of "struct
memberlist" structures was declared that is referenced from a type's
tp_members slot to change the type of the structure to PyMemberDef;
"struct memberlist" is now only used by old code that still calls
PyMember_Get/Set. The code in PyObject_GenericGetAttr/SetAttr now
calls the new APIs PyMember_GetOne/SetOne, which take a PyMemberDef
argument.
As examples, I added actual docstrings to the attributes of a few
types: file, complex, instance method, super, and xxsubtype.spamlist.
Also converted the symtable to new style getattr.
com_factor(): when a unary minus is attached to a float or imaginary zero,
don't optimize the UNARY_MINUS opcode away: the const dict can't
distinguish between +0.0 and -0.0, so ended up treating both like the
first one added to it. Optimizing UNARY_PLUS away isn't a problem.
(BTW, I already uploaded the 2.2a3 Windows installer, and this isn't
important enough to delay the release.)
Neal Norwitz
Nick Coghlan
Neil Schemenauer
Armin Rigo
Jeremy Hylton
Tim Peters
Phillip J. Eby
Brett Cannon
Raymond Hettinger
Anthony Baxter
Michael W. Hudson
Martin v. Löwis
Guido van Rossum
Just van Rossum
Gustavo Niemeyer
Walter Dörwald
Barry Warsaw
Andrew MacIntyre
Skip Montanaro
Marc-André Lemburg
Fred Drake
Sjoerd Mullender