Commit Graph

170 Commits

Author SHA1 Message Date
Serhiy Storchaka c86ca26d32 Issue #23096: Pickle representation of floats with protocol 0 now is the same
for both Python and C implementations.
2015-02-15 14:18:32 +02:00
Serhiy Storchaka 707b5ccde5 Issue #22783: Pickling now uses the NEWOBJ opcode instead of the NEWOBJ_EX
opcode if possible.
2014-12-16 19:43:46 +02:00
Serhiy Storchaka 01bdd9a980 Issue #15513: Added a __sizeof__ implementation for pickle classes. 2014-12-16 19:40:58 +02:00
Serhiy Storchaka 5bbd231f27 Issue #15513: Added a __sizeof__ implementation for pickle classes. 2014-12-16 19:39:08 +02:00
Antoine Pitrou 6cd5eda0dd Fix uninitialized variable after #22676. 2014-12-02 00:20:03 +01:00
Antoine Pitrou fce60eaf15 Issue #22676: Make the pickling of global objects which don't have a __module__ attribute less slow. 2014-10-23 22:47:50 +02:00
Serhiy Storchaka 67c719b34b Silenced some warnings about comparison between signed and unsigned integer
expressions.
2014-09-05 10:10:23 +03:00
Victor Stinner 049e509a9f Issue #22207: Fix "comparison between signed and unsigned integers" warning in
test checking for integer overflow on Py_ssize_t type: cast explicitly to
size_t.
2014-08-17 22:20:00 +02:00
Victor Stinner 7270b7f1aa _pickle: Optimize raw_unicode_escape(), use directly a bytes object, don't use
a temporary bytearray object.
2014-08-17 21:14:46 +02:00
Victor Stinner f13c46cc69 Issue #22218: Fix "comparison between signed and unsigned integers" warnings in
Modules/_pickle.c.
2014-08-17 21:05:55 +02:00
Larry Hastings 581ee3618c Issue #20326: Argument Clinic now uses a simple, unique signature to
annotate text signatures in docstrings, resulting in fewer false
positives.  "self" parameters are also explicitly marked, allowing
inspect.Signature() to authoritatively detect (and skip) said parameters.

Issue #20326: Argument Clinic now generates separate checksums for the
input and output sections of the block, allowing external tools to verify
that the input has not changed (and thus the output is not out-of-date).
2014-01-28 05:00:08 -08:00
Serhiy Storchaka 3c1f0f1b42 Issue #20395: Extract generated clinic code in Modules/_pickle.c to separate file. 2014-01-27 10:34:22 +02:00
Christian Heimes 27ea78b352 silence compiler warning that 's' may be used uninitialized in the load function. 2014-01-27 01:03:53 +01:00
Larry Hastings c20472640c Issue #20390: Small fixes and improvements for Argument Clinic. 2014-01-25 20:43:29 -08:00
Larry Hastings 5c66189e88 Issue #20189: Four additional builtin types (PyTypeObject,
PyMethodDescr_Type, _PyMethodWrapper_Type, and PyWrapperDescr_Type)
have been modified to provide introspection information for builtins.
Also: many additional Lib, test suite, and Argument Clinic fixes.
2014-01-24 06:17:25 -08:00
Larry Hastings 462582651c Two minor Argument Clinic bugfixes: use the name of the class in the
docstring for __new__ and __init__, and always use "goto exit" instead of
returning "NULL" for failure to parse (as _new__ and __init__ return ints).
2014-01-22 03:05:49 -08:00
Larry Hastings b7ccb20423 Issue #20294: Argument Clinic now supports argument parsing for __new__ and
__init__ functions.
2014-01-18 23:50:21 -08:00
Larry Hastings bebf73511a Issue #20287: Argument Clinic's output is now configurable, allowing
delaying its output or even redirecting it to a separate file.
2014-01-17 17:47:17 -08:00
Larry Hastings 8666e65206 Issue #20228: Argument Clinic now has special support for class special
methods.
2014-01-12 14:12:59 -08:00
Larry Hastings 61272b77b0 Issue #19273: The marker comments Argument Clinic uses have been changed
to improve readability.
2014-01-07 12:41:53 -08:00
Larry Hastings 3cceb38486 Issue #19976: Argument Clinic METH_NOARGS functions now always
take two parameters.
2014-01-04 11:09:09 -08:00
Stefan Krah f483b0f3a9 Issue #19972: Add rarely used freefunc. This fixes a leak if sys.exit()
is used in a program.
2013-12-14 13:43:10 +01:00
Alexandre Vassalotti d05c9ff845 Issue #6784: Strings from Python 2 can now be unpickled as bytes objects.
Initial patch by Merlijn van Deen.

I've added a few unrelated docstring fixes in the patch while I was at
it, which makes the documentation for pickle a bit more consistent.
2013-12-07 01:09:27 -08:00
Alexandre Vassalotti 6e73ff1a31 Issue #19881: Fix bad pickling of large bytes in cpickle. 2013-12-05 19:29:32 -08:00
Alexandre Vassalotti 2ccf8e969c Issue #6477: Merge with 3.3. 2013-11-30 17:58:53 -08:00
Alexandre Vassalotti 65846c6c51 Issue #6477: Keep PyNotImplemented_Type and PyNone_Type private. 2013-11-30 17:55:48 -08:00
Alexandre Vassalotti 3c23e7a5dc Issue #6477: Merge with 3.3. 2013-11-30 16:21:20 -08:00
Alexandre Vassalotti 19b6fa6ebb Issue #6477: Added support for pickling the types of built-in singletons. 2013-11-30 16:06:39 -08:00
Alexandre Vassalotti a1eedf9ff0 Merge with 3.3. 2013-11-30 13:55:39 -08:00
Alexandre Vassalotti 896414fedf Fixed _pickle.Unpickler to handle empty persistent IDs correctly. 2013-11-30 13:52:35 -08:00
Serhiy Storchaka 6fe39b76a9 Issue #17897: Optimized unpickle prefetching. 2013-11-30 23:15:38 +02:00
Alexandre Vassalotti 567eba1852 Use PyDict_GetItemWithError instead of PyDict_GetItem in cpickle. 2013-11-28 17:09:16 -08:00
Alexandre Vassalotti 6bf41e54a4 Remove explicit empty tuple reuse in cpickle.
PyTuple_New(0) always returns the same empty tuple from its free list anyway,
so we are not saving much here. Plus, the code where this was used is on
uncommon run paths.
2013-11-28 15:17:29 -08:00
Alexandre Vassalotti b13e6bcbd8 Remove the tuple reuse optimization in _Pickle_FastCall.
I have noticed a race-condition occurring on one of the buildbots because of
this optimization. The function called may release the GIL which means
multiple threads may end up accessing the shared tuple. I could fix it up by
storing the tuple to the Pickler and Unipickler object again, but honestly it
really not worth the trouble.

I ran many benchmarks and the only time the optimization helps is when using a
fin-memory file, like io.BytesIO on which reads are super cheap, combined with
pickle protocol less than 4. Even in this contrived case, the speedup is a
about 5%. For everything else, this optimization does not provide any
noticable improvements.
2013-11-28 14:56:09 -08:00
Alexandre Vassalotti 23bdd83053 Encapsulate cpickle global state in a dedicated object.
This implements PEP 3121 module finalization as well. This change does not
cause any significant impact on performance.
2013-11-27 19:36:52 -08:00
Alexandre Vassalotti 20c28c1ea2 Combine the FastCall functions in cpickle.
I fixed the bug that was in my previous attempt of this cleanup. I ran
the full test suite to verify I didn't introduce any obvious bugs.
2013-11-27 02:26:54 -08:00
Alexandre Vassalotti b4a04fb7e6 Reverting e39db21df580 eagerly due to buildbot failures. 2013-11-25 13:25:12 -08:00
Alexandre Vassalotti f94a041f8f Combine _Pickler_FastCall and _Unpickler_FastCall in cpickle. 2013-11-25 13:03:32 -08:00
Alexandre Vassalotti 1048fb5539 Issue #19739: Try to fix compiler warnings on 32-bit Windows. 2013-11-25 11:35:46 -08:00
Alexandre Vassalotti ded929b300 Merge save_int into save_long in cpickle to remove redundant code.
Also, replace unnessary uses of the #if preprocessor directive.
2013-11-24 22:41:13 -08:00
Alexandre Vassalotti 8a67f52463 Simplify save_bool in cpickle. 2013-11-24 21:40:18 -08:00
Alexandre Vassalotti ed8c906127 Use Clinic to process arguments in cpickle.
This doesn't make any functional changes to the exisiting implementation.  The
conversion did help however uncover documentation bugs. The best thing about
this conversion is less C code to maintain by hand.
2013-11-24 12:25:48 -08:00
Alexandre Vassalotti fc91285c93 Remove code path in cpickle that does not exist in pickle. 2013-11-24 03:07:35 -08:00
Alexandre Vassalotti c49477b184 Make Ellipsis and NotImplemented picklable through the reduce protocol. 2013-11-24 02:53:45 -08:00
Alexandre Vassalotti 4c05d3bc56 Make built-in methods picklable through the reduce protocol. 2013-11-24 02:41:05 -08:00
Alexandre Vassalotti b6a2f2a0d1 Make framing optional in pickle protocol 4.
This will allow us to control in the future whether to use framing or not.
For example, we may want to turn it off for tiny pickle where it doesn't
help.

The change also improves performance slightly:

### fastpickle ###
Min: 0.608517 -> 0.557358: 1.09x faster
Avg: 0.798892 -> 0.694738: 1.15x faster
Significant (t=3.45)
Stddev: 0.17145 -> 0.12704: 1.3496x smaller
Timeline: http://goo.gl/3xQE1J

### pickle_dict ###
Min: 0.669920 -> 0.615271: 1.09x faster
Avg: 0.733633 -> 0.645058: 1.14x faster
Significant (t=5.05)
Stddev: 0.12041 -> 0.02961: 4.0662x smaller
Timeline: http://goo.gl/LpLSXI

### pickle_list ###
Min: 0.397583 -> 0.368112: 1.08x faster
Avg: 0.412784 -> 0.397223: 1.04x faster
Significant (t=2.78)
Stddev: 0.01518 -> 0.03653: 2.4068x larger
Timeline: http://goo.gl/v39E59

### unpickle_list ###
Min: 0.692935 -> 0.594870: 1.16x faster
Avg: 0.730012 -> 0.628395: 1.16x faster
Significant (t=17.76)
Stddev: 0.02720 -> 0.02995: 1.1012x larger
Timeline: http://goo.gl/2P9AEt

The following not significant results are hidden, use -v to show them:
fastunpickle.
2013-11-23 20:30:03 -08:00
Antoine Pitrou c1207c1bcf Fix signed / unsigned comparison 2013-11-23 21:34:04 +01:00
Gregory P. Smith 2b38fc187c gcc doesn't realize that dummy is always initialized by the function call
and warns about potential uninitialized use.
Silence that by initializing it to null.
2013-11-23 20:21:28 +00:00
Christian Heimes e8b1ba1699 Issue #17810: Add two missing error checks to save_global
CID 1131946:  Unchecked return value  (CHECKED_RETURN)
2013-11-23 21:13:39 +01:00
Antoine Pitrou 6188d09f1c Merge 2013-11-23 21:06:21 +01:00