- tuplecontains(): call RichCompare(Py_EQ).
- Get rid of tuplecompare(), in favor of new tuplerichcompare() (a
clone of list_compare()).
- Aligned the comments for large struct initializers.
earlier coercion changes, not by rich comparisons. When a coercion
function returns 1 (meaning it cannot do it), it should not INCREF the
arguments. When no __coerce__() method was found, instance_coerce()
originally returned 0, pretending it did it. Neil changed the return
value to 1, more accurately reflecting that it didn't do anything, but
forgot to take out the two INCREF calls.
- sort's docompare() calls RichCompare(Py_LT).
- list_contains(), list_index(), listcount(), listremove() call
RichCompare(Py_EQ).
- Get rid of list_compare(), in favor of new list_richcompare(). The
latter does some nice shortcuts, like when == or != is requested, it
first compares the lengths for trivial accept/reject. Then it goes
over the items until it finds an index where the items differe; then
it does more shortcut magic to minimize the number of additional
comparisons.
- Aligned the comments for large struct initializers.
- Use the compare nesting level and in-progress dictionary properly in
PyObject_RichCompare().
- Change the in-progress code to use static variables instead of
globals (both the nesting level and the key for the thread dict were
globals but have no reason to be globals; the key can even be a
function-static variable in get_inprogress_dict()).
- Rewrote try_rich_to_3way_compare() to benefit from the similarity of
the three cases, making it table-driven.
- In try_rich_to_3way_compare(), test for EQ before LT and GT. This
turns out essential when comparing recursive UserList instances;
with the old code, these would recurse into rich comparison three
times for each nesting level up to NESTING_LIMIT/2, making the total
number of calls in the order of 3**(NESTING_LIMIT/2)!
NOTE: I'm not 100% comfortable with this. It works for the standard
test suite (which compares a few trivial recursive data structures
only), but I'm not sure that the in-progress dictionary is used
properly by the rich comparison code. Jeremy suggested that maybe the
operation should be included in the dict. Currently I presume that
objects in the dict are equal unless proven otherwise, and I set the
outcome for the rich comparison accordingly: true for operators EQ,
LE, GE, and false for the other three. But Jeremy seems to think that
there may be counter-examples where this doesn't do the right thing.
except that it always returns Unicode objects.
A new C API PyObject_Unicode() is also provided.
This closes patch #101664.
Written by Marc-Andre Lemburg. Copyright assigned to Guido van Rossum.
- Got rid of instance_cmp(); refactored instance_compare().
- Added instance_richcompare() which calls __lt__() etc.
Some unrelated stuff mixed in:
- Aligned comments in various large struct initializers.
- Better test to avoid recursion if __coerce__ returns self as the
first argument (this is an unrelated fix by Neil Schemenauer!).
- Style nit: don't use Py_DECREF(Py_NotImplemented); use
Py_DECREF(result) -- it just looks better. :-)
PyObject_RichCompare() and PyObject_RichCompareBool().
XXX Note: the code that checks for deeply nested rich comparisons is
bogus -- it assumes the two objects are always identical, rather than
using the same logic as PyObject_Compare(). I'll fix that later.
simpler if we use fgetpos and fsetpos, rather than trying to mess with
platform-specific TELL64 alternatives.
Of course, this hasn't been tested on a 64-bit platform, so I may have
to withdraw this -- but I'm hopeful, and Trent Mick supports this
patch!
in case the parameters are out of bounds and fixes error handling
for .count(), .startswith() and .endswith() for the case of
mixed string/Unicode objects.
This patch adds Python style index semantics to PyUnicode_Count()
indices (including the special handling of negative indices).
The patch is an extended version of patch #103249 submitted
by Michael Hudson (mwh) on SF. It also includes new test cases.
Closes SF patch #103123.
funcobject.h:
PyFunctionObject: add the func_dict slot.
funcobject.c:
PyFunction_New(): Initialize the func_dict slot to NULL.
func_getattr(): Rename to func_getattro() and change the
signature. It's more efficient to use attro methods and dig the C
string out than it is to re-convert a C string to a PyString.
Also, add support for getting the __dict__ (a.k.a. func_dict)
attribute, and for getting an arbitrary function attribute.
func_setattr(): Rename to func_setattro() and change the signature
for the same reason. Also add support for setting __dict__
(a.k.a. func_dict) and any arbitrary function attribute.
func_dealloc(): Be sure to DECREF the func_dict slot.
func_traverse(): Be sure to traverse func_dict too.
PyFunction_Type: make the necessary func_?etattro() changes.
classobject.c:
instancemethod_memberlist: Add __dict__
instancemethod_setattro(): New method to set arbitrary attributes
on methods (really the underlying im_func). Raise TypeError when
the instance is bound or when you're trying to set one of the
reserved im_* attributes.
instancemethod_getattr(): Renamed to instancemethod_getattro()
since that's what it really is. Also, added support fo getting
arbitrary attributes through the im_func.
PyMethod_Type: Do the ?etattr{,o} dance.
object.
This fixes potential overflows in xrange()'s internal calculations on
64-bit platforms. The fix is complicated because the sq_length slot
function can only return an int; we want to support
xrange(sys.maxint), which is a 64-bit quantity on most 64-bit
platforms (except Win64). The solution is hacky but the best
possible: when the range is that long, we can use it in a for loop but
we can't ask for its length (nor can we actually iterate beyond
2**31-1, because the sq_item slot function has the same restrictions
on its arguments. Fixing those restrictions is a project for another
day...
faster than the other. Should be faster for Mark Favas's 254-character
mail log lines, and *is* 3-4% quicker for my test case with much shorter
lines (but they're typical of *my* text files, and I'm tired of optimizing
for everyone else at my expense <wink> -- in fact, the only one who loses
here is Guido ...).
Tim discovered another "bug" in my get_line() code: while the comments
said that n<0 was invalid, it was in fact still called with n<0 (when
PyFile_GetLine() was called with n<0). In that case fortunately
executed the same code as for n==0.
Changed the comment to admit this fact, and changed Tim's MS speed
hack code to use 'n <= 0' as the criteria for the speed hack.
code duplication is to let us get away without a realloc whenever possible;
boosted the init buf size (the cutoff at which we *can* get away without
a realloc) from 100 to 200 so that more files can enjoy this boost; and
allowed other threads to run in all cases. The last two cost something,
but not significantly: in my fat test case, less than a 1% slowdown total.
Since my test case has a great many short lines, that's probably the worst
slowdown, too. While the logic barely changed, there were lots of edits.
This also gets rid of the reference to fp->_cnt, so the last platform
assumption being made here is that fgets doesn't overwrite bytes
capriciously (== beyond the terminating null byte it must write).
variant that never needs to "search from the right".
Also fixed unlikely memory leak in get_line, if string size overflows INTMAX.
Also new std test test_bufio to make sure .readline() works.
realized that this behavior is already present in PyFile_GetLine(),
which is the only place that needs it. A little refactoring of that
function made get_line_raw() redundant.
The mapping dictionaries can now contain 1-n mappings, meaning
that character ordinals may be mapped to strings or Unicode object,
e.g. 0x0078 ('x') -> u"abc", causing the ordinal to be replaced by
the complete string or Unicode object instead of just one character.
Another feature introduced by the patch is that of mapping oridnals to
the emtpy string. This allows removing characters.
The patch is different from patch #103100 in that it does not cause a
performance hit for the normal use case of 1-1 mappings.
Written by Marc-Andre Lemburg, copyright assigned to Guido van Rossum.
- The raw_input() functionality is moved to a separate function.
- Drop GNU getline() in favor of getc_unlocked(), which exists on more
platforms (and is even a tad faster on my system).
codec to not apply Latin-1 mappings for keys which are not found
in the mapping dictionaries, but instead treat them as undefined
mappings.
The patch was originally written by Martin v. Loewis with some
additional (cosmetic) changes and an updated test script
by Marc-Andre Lemburg.
The standard codecs were recreated from the most current files
available at the Unicode.org site using the Tools/scripts/gencodec.py
tool.
This patch closes the bugs #116285 and #119960.
raise ValueError. Checked in the patch as far as it went, but also changed
all of ints, longs and floats to raise ZeroDivisionError instead when raising
0 to a negative number. This is what 754-inspired stds require, as the "true
result" is an infinity obtained from finite operands, i.e. it's a singularity.
Also changed float pow to not be so timid about using its square-and-multiply
algorithm. Note that what math.pow does is unrelated to what builtin pow
does, and will still vary by platform.
result-object-pointer that is passed in, when an exception occurs during
coercion. The pointer has to be explicitly initialized in the caller to avoid
putting trash on the Python stack.
#define'd to an unreasonable value (several recent gcc systems have
misdefined it, causing bogus overflows in integer multiplication). Nuke
CHAR_BIT entirely.
after unicode_empty has been freed, otherwise it might not point to
the real start of the unicode_freelist. Final closure for SF bug
#110681, Jitterbug PR#398.
Add definitions of INT_MAX and LONG_MAX to pyport.h.
Remove includes of limits.h and conditional definitions of INT_MAX
and LONG_MAX elsewhere.
This closes SourceForge patch #101659 and bug #115323.
- use unidb compression for the unicodectype module. smaller, faster,
and slightly more portable...
(note: this commit doesn't include the unicodectype.c file itself; I'm
still waiting for the reviewers...)
I fixed the specific complaint but left the (many) large issues untouched.
See the (very long) bug report discussion for why:
http://sourceforge.net/bugs/?func=detailbug&group_id=5470&bug_id=110624
Note that while I left the interface to the undocumented public API function
PyFloat_FromString alone, its 2nd argument is useless. From a comment block
in the code:
RED_FLAG 22-Sep-2000 tim
PyFloat_FromString's pend argument is braindead. Prior to this RED_FLAG,
1. If v was a regular string, *pend was set to point to its terminating
null byte. That's useless (the caller can find that without any
help from this function!).
2. If v was a Unicode string, or an object convertible to a character
buffer, *pend was set to point into stack trash (the auto temp
vector holding the character buffer). That was downright dangerous.
Since we can't change the interface of a public API function, pend is
still supported but now *officially* useless: if pend is not NULL,
*pend is set to NULL.
Note a curious extension to the std C rules: x, X and o formatting can never produce
a sign character in C, so the '+' and ' ' flags are meaningless for them. But
unbounded ints *can* produce a sign character under these conversions (no fixed-
width bitstring is wide enough to hold all negative values in 2's-comp form). So
these flags become meaningful in Python when formatting a Python long which is too
big to fit in a C long. This required shuffling around existing code, which hacked
x and X conversions to death when both the '#' and '0' flags were specified: the
hacks weren't strong enough to deal with the simultaneous possibility of the ' ' or
'+' flags too, since signs were always meaningless before for x and X conversions.
Isomorphic shuffling was required in unicodeobject.c.
Also added dozens of non-trivial new unbounded-int test cases to test_format.py.
which implements the automatic conversion from Unicode to a string
object using the default encoding.
The new API is then put to use to have eval() and exec accept
Unicode objects as code parameter. This closes bugs #110924
and #113890.
As side-effect, the traditional C APIs PyString_Size() and
PyString_AsString() will also accept Unicode objects as
parameters.
objects for the attribute name. Unicode objects are converted to
a string using the default encoding before trying the lookup.
Note that previously it was allowed to pass arbitrary objects as
attribute name in case the tp_getattro/setattro slots were defined.
This patch fixes this by applying an explicit string check first:
all uses of these slots expect string objects and do not check
for the type resulting in a core dump. The tp_getattro/setattro
are still useful as optimization for lookups using interned
string objects though.
This patch fixes bug #113829.
that Py_INCREF boosts global _Py_RefTotal when Py_REF_DEBUG is defined
but Py_TRACE_REFS isn't.
There are, IMO, way too many preprocessor gimmicks in use for refcount
debugging (at least 3 distinct true/false symbols, but not all 8 combos
are supported by the code, etc etc), and no coherent documentation of
this stuff -- 'twas too painful to track this one down.
all, either to see whether the # of chars fit in an int, or that the
amount of memory needed fit in a size_t. Checking these is expensive, but
the alternative is silently wrong answers (as in the bug report) or
core dumps (which were easy to provoke using Unicode strings).
exception context. This avoids improperly propogating errors raised by
a user-defined __cmp__() by a subsequent lookup operation.
This patch does *not* include the performance enhancement patch for
dictionaries with string keys only; that will be checked in separately.
This closes SourceForge patch #101277 and bug #112558.
file.writelines() now tries to emulate the behaviour of file.write()
as closely as possible. Due to the problems with releasing the
interpreter lock the solution isn't exactly optimal, but still better
than not supporting the file.write() semantics at all.
types (i.e. Py_uintptr_t, our spelling of C9X's uintptr_t). ANSI
specifies that pointer compares other than == and != to non-related
structures are undefined. This quiets an Insure portability warning.
scope. Previously, s_buffer[] was defined inside the
PyUnicode_Check() scope, but referred to in the outer scope via
assignment to s. This quiets an Insure portability warning.
to integer types (i.e. Py_uintptr_t, our spelling of C9X's uintptr_t).
ANSI specifies that pointer compares other than == and != to
non-related structures are undefined. This quiets an Insure
portability warning.
is no __getslice__ available. Also does the same for C extension types.
Includes rudimentary documentation (it could use a cross reference to the
section on slice objects, I couldn't figure out how to do that) and a test
suite for all Python __hooks__ I could think of, including the new
behaviour.
shutdown time, but CVS log entry for revision 2.45 explains why this
is so. Simply include a comment so we don't have to re-figure it out
again 5 years from now.
This was a misleading bug -- the true "bug" was that hash(x) gave an error
return when x is an infinity. Fixed that. Added new Py_IS_INFINITY macro to
pyport.h. Rearranged code to reduce growing duplication in hashing of float and
complex numbers, pushing Trent's earlier stab at that to a logical conclusion.
Fixed exceedingly rare bug where hashing of floats could return -1 even if there
wasn't an error (didn't waste time trying to construct a test case, it was simply
obvious from the code that it *could* happen). Improved complex hash so that
hash(complex(x, y)) doesn't systematically equal hash(complex(y, x)) anymore.
resized after creation. 0-length strings are usually shared
and _PyString_Resize() fails on these shared strings.
Fixes [ Bug #111667 ] unicode core dump.
Properly end a comment block. It was terminated fine later but by a subsequent
block and. It was also in #if 0. This patch is so trivial I can't believe I am
talking about it. :)
function (together with other locale aware ones) should into a new collation
support module. See python-dev for a discussion of this removal.
Note: This patch should also be applied to the 1.6 branch.
the Python Unicode implementation.
The internal buffer used for implementing the buffer protocol
is renamed to defenc to make this change visible. It now holds the
default encoded version of the Unicode object and is calculated
on demand (NULL otherwise).
Since the default encoding defaults to ASCII, this will mean that
Unicode objects which hold non-ASCII characters will no longer
work on C APIs using the "s" or "t" parser markers. C APIs must now
explicitly provide Unicode support via the "u", "U" or "es"/"es#"
parser markers in order to work with non-ASCII Unicode strings.
(Note: this patch will also have to be applied to the 1.6 branch
of the CVS tree.)
This doesn't change the copyright status for these files -- just the
markings! Doing it on the main branch for these three files for which
the HEAD revision was pushed back into 1.6.
The UTF-8 decoder is still buggy (i.e. it doesn't pass Markus Kuhn's
stress test), mainly due to the following construct:
#define UTF8_ERROR(details) do { \
if (utf8_decoding_error(&s, &p, errors, details)) \
goto onError; \
continue; \
} while (0)
(The "continue" statement is supposed to exit from the outer loop,
but of course, it doesn't. Indeed, this is a marvelous example of
the dangers of the C programming language and especially of the C
preprocessor.)
comments, docstrings or error messages. I fixed two minor things in
test_winreg.py ("didn't" -> "Didn't" and "Didnt" -> "Didn't").
There is a minor style issue involved: Guido seems to have preferred English
grammar (behaviour, honour) in a couple places. This patch changes that to
American, which is the more prominent style in the source. I prefer English
myself, so if English is preferred, I'd be happy to supply a patch myself ;)
use PyString_AS_STRING macro on local string object
when resizing string, make sure resized string will always be big enough
split string containing error message across two lines
add test to string_tests that causes resizing
seqlen==1 clause, before returning item, we need to DECREF seq. In
the res=PyString... failure clause, we need to goto finally to also
decref seq (and the DECREF of res in finally is changed to a
XDECREF). Also, we need to DECREF seq just before the
PyUnicode_Join() return.
implementation -- use PySequence_Fast interface to iterate over elements
interface -- if instance object reports wrong length, ignore it;
previous version raised an IndexError if reported length was too high
value is calculated from the character values, in a way
that makes sure an 8-bit ASCII string and a unicode string
with the same contents get the same hash value.
(as a side effect, this also works for ISO Latin 1 strings).
for more details, see the python-dev discussion.
was cascades of warnings about mismatching const decls. Overall,
I think const creates lots of headaches and solves almost
nothing. Added enough consts to shut up the warnings, but
this did require casting away const in one spot too (another
usual outcome of starting down this path): the function
mymemreplace can't return const char*, but sometimes wants to
return its first argument as-is, which latter must be declared
const char* in order to avoid const warnings at mymemreplace's
call sites. So, in the case the function wants to return the
first arg, that arg's declared constness must be subverted.
This was a convenient excuse to create the pyport.h file recently
discussed!
Please use new Py_ARITHMETIC_RIGHT_SHIFT when right-shifting a
signed int and you *need* sign-extension. This is #define'd in
pyport.h, keying off new config symbol SIGNED_RIGHT_SHIFT_ZERO_FILLS.
If you're running on a platform that needs that symbol #define'd,
the std tests never would have worked for you (in particular,
at least test_long would have failed).
The autoconfig stuff got added to Python after my Unix days, so
I don't know how that works. Would someone please look into doing
& testing an auto-config of the SIGNED_RIGHT_SHIFT_ZERO_FILLS
symbol? It needs to be defined if & only if, e.g., (-1) >> 3 is
not -1.
Stein -- thanks!). Incidentally removed all the Py_PROTO macros
from object.h, as they prevented my editor from magically finding
the definitions of the "coercion", "cmpfunc" and "reprfunc"
typedefs that were being redundantly applied in longobject.c.
works just like the Unicode one. The C APIs match the ones in the Unicode
implementation, but were extended to be able to reuse the existing
Unicode codecs for string purposes too.
Conversions from string to Unicode and back are done using the
default encoding.
implementation. This was really to test whether my new CVS+SSH
setup is more usable than the old one -- and turns out it is (for
whatever reason, it was impossible to do a commit before that
involved more than one directory).
The common technique for printing out a pointer has been to cast to a long
and use the "%lx" printf modifier. This is incorrect on Win64 where casting
to a long truncates the pointer. The "%p" formatter should be used instead.
The problem as stated by Tim:
> Unfortunately, the C committee refused to define what %p conversion "looks
> like" -- they explicitly allowed it to be implementation-defined. Older
> versions of Microsoft C even stuck a colon in the middle of the address (in
> the days of segment+offset addressing)!
The result is that the hex value of a pointer will maybe/maybe not have a 0x
prepended to it.
Notes on the patch:
There are two main classes of changes:
- in the various repr() functions that print out pointers
- debugging printf's in the various thread_*.h files (these are why the
patch is large)
Closes SourceForge patch #100505.
errors in some of the hash algorithms. For exmaple, in float_hash and
complex_hash a certain part of the value is not included in the hash
calculation. See Tim's, Guido's, and my discussion of this on
python-dev in May under the title "fix float_hash and complex_hash for
64-bit *nix"
(2) The hash algorithms that use pointers (e.g. func_hash, code_hash)
are universally not correct on Win64 (they assume that sizeof(long) ==
sizeof(void*))
As well, this patch significantly cleans up the hash code. It adds the
two function _Py_HashDouble and _PyHash_VoidPtr that the various
hashing routine are changed to use.
These help maintain the hash function invariant: (a==b) =>
(hash(a)==hash(b))) I have added Lib/test/test_hash.py and
Lib/test/output/test_hash to test this for some cases.
Avoid calling the dealloc function, previously triggered with
DECREF(inst). This caused a segfault in PyDict_GetItem, called with a
NULL dict, whenever inst->in_dict fails under low-memory conditions.
This patch modifies the type structures of objects that
participate in GC. The object's tp_basicsize is increased when
GC is enabled. GC information is prefixed to the object to
maintain binary compatibility. GC objects also define the
tp_flag Py_TPFLAGS_GC.
Fixed a bug in PyUnicode_Count() which would have caused a
core dump in case of substring coercion failure.
Synchronized .count() with the string method of the same name
to return len(s)+1 for s.count('').
The following patch adds "sq_contains" support to rangeobject, and enables
the already-written support for sq_contains in listobject and tupleobject.
The rangeobject "contains" code should be a bit more efficient than the
current default "in" implementation ;-) It might not get used much, but it's
not that much to add.
listobject.c and tupleobject.c already had code for sq_contains, and the
proper struct member was set, but the PyType structure was not extended to
include tp_flags, so the object-specific code was not getting called (Go
ahead, test it ;-). I also did this for the immutable_list_type in
listobject.c, eventhough it is probably never used. Symmetry and all that.
Fixed %c formatting to check for one character arguments. Thanks
to Finn Bock for finding this bug.
Added a fix for bug PR#348 which originated from not resetting
the globals correctly in _PyUnicode_Fini().
Change the default encoding to 'ascii' (it was previously
defined as UTF-8).
Note: The implementation still uses UTF-8 to implement
the buffer protocol, so C APIs will still see UTF-8. This
is on purpose: rather than fixing the Unicode implementation,
the C APIs should be made Unicode aware.
This patch correct bounds checking in PyLong_FromLongLong. Currently, it does
not check properly for negative values when checking to see if the incoming
value fits in a long or unsigned long. This results in possible silent
truncation of the value for very large negative values.
Added support for user settable default encodings. The
current implementation uses a per-process global which
defines the value of the encoding parameter in case it
is set to NULL (meaning: use the default encoding).
Fix the string methods that implement slice-like semantics with
optional args (count, find, endswith, etc.) to properly handle
indeces outside [INT_MIN, INT_MAX]. Previously the "i" formatter
for PyArg_ParseTuple was used to get the indices. These could overflow.
This patch changes the string methods to use the "O&" formatter with
the slice_index() function from ceval.c which is used to do the same
job for Python code slices (e.g. 'abcabcabc'[0:1000000000L]).
Fix the string methods that implement slice-like semantics with
optional args (count, find, endswith, etc.) to properly handle
indeces outside [INT_MIN, INT_MAX]. Previously the "i" formatter
for PyArg_ParseTuple was used to get the indices. These could overflow.
This patch changes the string methods to use the "O&" formatter with
the slice_index() function from ceval.c which is used to do the same
job for Python code slices (e.g. 'abcabcabc'[0:1000000000L]). slice_index()
is renamed _PyEval_SliceIndex() and is now exported. As well, the return
values for success/fail were changed to make slice_index directly
usable as required by the "O&" formatter.
[GvR: shouldn't a similar patch be applied to unicodeobject.c?]
gave bogus results for chars in the range 128-255, because their
implementation was using signed characters. Fixed this by using
unsigned character pointers (as opposed to using Py_CHARMASK()).
For more comments, read the patches@python.org archives.
For documentation read the comments in mymalloc.h and objimpl.h.
(This is not exactly what Vladimir posted to the patches list; I've
made a few changes, and Vladimir sent me a fix in private email for a
problem that only occurs in debug mode. I'm also holding back on his
change to main.c, which seems unnecessary to me.)
Fixed a reference leak in the allocator.
Renamed utf8_string to _PyUnicode_AsUTF8String() and made
it external for use by other parts of the interpreter.
The previous checkin (2.84) added a PyErr_Format call that made the
cost of raising an AttributeError much more expensive. In general
this doesn't matter, except that checks for __init__ and
__del__ methods, where exceptions are caught and cleared in C, also
got much more expensive.
The fix is to split instance_getattr1 into two calls:
instance_getattr2 checks the instance and the class for the attribute
and returns it or returns NULL on error. It does not raise an
exception.
instance_getattr1 does rexec checks, then calls instance_getattr2. It
raises an exception if instance_getattr2 returns NULL.
PyInstance_New and instance_dealloc now call instance_getattr2
directly.
Improvements:
- does no longer need any extra memory
- has no relationship to tstate
- works in debug mode
- can easily be modified for free threading (hi Greg:)
Side effects:
Trashcan does change the order of object destruction.
Prevending that would be quite an immense effort, as
my attempts have shown. This version works always
the same, with debug mode or not. The slightly
changed destruction order should therefore be no problem.
Algorithm:
While the old idea of delaying the destruction of some
obejcts at a certain recursion level was kept, we now
no longer aloocate an object to hold these objects.
The delayed objects are instead chained together
via their ob_type field. The type is encoded via
ob_refcnt. When it comes to the destruction of the
chain of waiting objects, the topmost object is popped
off the chain and revived with type and refcount 1,
then it gets a normal Py_DECREF.
I am confident that this solution is near optimum
for minimizing side effects and code bloat.
_PyTuple_Resize(). In addition, a change suggested by Jeremy Hylton
to limit the size of the free lists is also merged into this patch.
Charles wrote initially:
"""
Test Case: run the following code:
class Nothing:
def __len__(self):
return 5
def __getitem__(self, i):
if i < 3:
return i
else:
raise IndexError, i
def g(a,*b,**c):
return
for x in xrange(1000000):
g(*Nothing())
and watch Python's memory use go up and up.
Diagnosis:
The analysis begins with the call to PySequence_Tuple at line 1641 in
ceval.c - the argument to g is seen to be a sequence but not a tuple,
so it needs to be converted from an abstract sequence to a concrete
tuple. PySequence_Tuple starts off by creating a new tuple of length
5 (line 1122 in abstract.c). Then at line 1149, since only 3 elements
were assigned, _PyTuple_Resize is called to make the 5-tuple into a
3-tuple. When we're all done the 3-tuple is decrefed, but rather than
being freed it is placed on the free_tuples cache.
The basic problem is that the 3-tuples are being added to the cache
but never picked up again, since _PyTuple_Resize doesn't make use of
the free_tuples cache. If you are resizing a 5-tuple to a 3-tuple and
there is already a 3-tuple in free_tuples[3], instead of using this
tuple, _PyTuple_Resize will realloc the 5-tuple to a 3-tuple. It
would more efficient to use the existing 3-tuple and cache the
5-tuple.
By making _PyTuple_Resize aware of the free_tuples (just as
PyTuple_New), we not only save a few calls to realloc, but also
prevent this misbehavior whereby tuples are being added to the
free_tuples list but never properly "recycled".
"""
And later:
"""
This patch replaces my submission of Sun, 16 Apr and addresses Jeremy
Hylton's suggestions that we also limit the size of the free tuple
list. I chose 2000 as the maximum number of tuples of any particular
size to save.
There was also a problem with the previous version of this patch
causing a core dump if Python was built with Py_TRACE_REFS. This is
fixed in the below version of the patch, which uses tupledealloc
instead of _Py_Dealloc.
"""
Guido van Rossum
Marc-André Lemburg
Barry Warsaw
Tim Peters
Neil Schemenauer
Jeremy Hylton
Fred Drake
Andrew M. Kuchling
Moshe Zadka
Fredrik Lundh
Thomas Wouters
Martin v. Löwis
Jack Jansen
Trent Mick
Vladimir Marangozov
Peter Schneider-Kamp
Greg Stein
Skip Montanaro
Sjoerd Mullender