It seems like this should be a different error than SystemError, but
I don't have any great ideas and SystemError was raised in 2.4 and earlier.
Will backport.
some warnings from Klokwork. They verify the assumptions of the format
of svn version output.
The assert in the thread module helped debug a problem on HP-UX.
* unified the way intobject, longobject and mystrtoul handle
values around -sys.maxint-1.
* in general, trying to entierely avoid overflows in any computation
involving signed ints or longs is extremely involved. Fixed a few
simple cases where a compiler might be too clever (but that's all
guesswork).
* more overflow checks against bad data in marshal.c.
* 2.5 specific: fixed a number of places that were still confusing int
and Py_ssize_t. Some of them could potentially have caused
"real-world" breakage.
* list.pop(x): fixing overflow issues on x was messy. I just reverted
to PyArg_ParseTuple("n"), which does the right thing. (An obscure
test was trying to give a Decimal to list.pop()... doesn't make
sense any more IMHO)
* trying to write a few tests...
The compiler was checking that there was something on the fblock
stack, but not that there was a loop on the stack. Fixed that and
added a test for the specific syntax error.
Bug fix candidate.
Building with HP's cc on HP-UX turned up a couple of problems.
_PyGILState_NoteThreadState was declared as static inconsistently.
Make it static as it's not necessary outside of this module.
Some tests failed because errno was reset to 0. (I think the tests
that failed were at least: test_fcntl and test_mailbox).
Ensure that errno doesn't change after a call to Py_END_ALLOW_THREADS.
This only affected debug builds.
generator expressions (x for x, in ... ) works again.
Sigh, I only fixed for loops the first time, not list comps and genexprs too.
I couldn't find any more unpacking cases where there is a similar bug lurking.
This code should be refactored to eliminate the duplication. I'm sure
the listcomp/genexpr code can be refactored. I'm not sure if the for loop
can re-use any of the same code though.
Will backport to 2.5 (the only place it matters).
I modified this patch some by fixing style, some error checking, and adding
XXX comments. This patch requires review and some changes are to be expected.
I'm checking in now to get the greatest possible review and establish a
baseline for moving forward. I don't want this to hold up release if possible.
However, there was no error checking that PyFloat_FromDouble returned
a valid pointer. I believe this change is correct as it seemed
to follow other code in the area.
Klocwork # 292.
there was no verification that privateobj was a PyString. If it wasn't
a string, this could have allowed a NULL pointer to creep in below and crash.
I wonder if this should be PyString_CheckExact? Must identifiers be strings
or can they be subclasses?
Klocwork #275
This is the first batch of fixes that should be easy to verify based on context.
This fixes problem numbers: 220 (ast), 323-324 (symtable),
321-322 (structseq), 215 (array), 210 (hotshot), 182 (codecs), 209 (etree).
PyThreadState_SetAsyncExc(): internal correctness changes wrt
refcount safety and deadlock avoidance. Also added a basic test
case (relying on ctypes) and repaired the docs.
on each iteration. I'm not positive this is the best way to handle
this. I'm also not sure that there aren't other cases where
the lnotab is generated incorrectly. It would be great if people
that use pdb or tracing could test heavily.
Also:
* Remove dead/duplicated code that wasn't used/necessary
because we already handled the docstring prior to entering the loop.
* add some debugging code into the compiler (#if 0'd out).
This provides the proper warning for struct.pack().
PyErr_Warn() is now deprecated in favor of PyErr_WarnEx().
As mentioned by Tim Peters on python-dev.
with PEP 302. This was fixed by adding an ``imp.NullImporter`` type that is
used in ``sys.path_importer_cache`` to cache non-directory paths and avoid
excessive filesystem operations during imports.
In general, C doesn't define anything about what happens when
an operation on a signed integral type overflows, and PyOS_strtol()
did several formally undefined things of that nature on signed
longs. Some version of gcc apparently tries to exploit that now,
and PyOS_strtol() could fail to detect overflow then.
Tried to repair all that, although it seems at least as likely to me
that we'll get screwed by bad platform definitions for LONG_MIN
and/or LONG_MAX now. For that reason, I don't recommend backporting
this.
Note that I have no box on which this makes a lick of difference --
can't really test it, except to note that it didn't break anything
on my boxes.
Silent change: PyOS_strtol() used to return the hard-coded 0x7fffffff
in case of overflow. Now it returns LONG_MAX. They're the same only on
32-bit boxes (although C doesn't guarantee that either ...).
Moved the code for _PyThread_CurrentFrames() up, so it's no longer
in a huge "#ifdef WITH_THREAD" block (I didn't realize it /was/ in
one).
Changed test_sys's test_current_frames() so it passes with or without
thread supported compiled in.
Note that test_sys fails when Python is compiled without threads,
but for an unrelated reason (the old test_exit() fails with an
indirect ImportError on the `thread` module). There are also
other unrelated compilation failures without threads, in extension
modules (like ctypes); at least the core compiles again.
Do we really support --without-threads? If so, there are several
problems remaining.
v2 can be NULL if exception2 is NULL. I don't think that condition can happen,
but I'm not sure it can't either. Now the code will protect against either
being NULL.
omit a default "error" argument for NULL pointer. This allows
the parser to take a codec from cjkcodecs again.
(Reported by Taewook Kang and reviewed by Walter Doerwald)
Heavily revised, comprising revisions:
46640 - original trunk revision (backed out in r46655)
46647 - markup fix (backed out in r46655)
46692:46918 merged from branch aimacintyre-sf1454481
branch tested on buildbots (Windows buildbots had problems
not related to these changes).
exact maximum size someone guesses is needed. In this case, if
we're really worried about extreme integers, then "cp%d" can
actually need 14 bytes (2 for "cp" + 1 for \0 at the end +
11 for -(2**31-1)). So reserve 128 bytes instead -- nothing is
actually saved by making a stack-local buffer tiny.
46640 Patch #1454481: Make thread stack size runtime tunable.
46647 Markup fix
The first is causing many buildbots to fail test runs, and there
are multiple causes with seemingly no immediate prospects for
repairing them. See python-dev discussion.
Note that a branch can (and should) be created for resolving these
problems, like
svn copy svn+ssh://svn.python.org/python/trunk -r46640 svn+ssh://svn.python.org/python/branches/NEW_BRANCH
followed by merging rev 46647 to the new branch.
set_exc_info(), reset_exc_info(): By exploiting the
likely (who knows?) invariant that when an exception's
`type` is NULL, its `value` and `traceback` are also NULL,
save some cycles in heavily-executed code.
This is a "a kronar saved is a kronar earned" patch: the
speedup isn't reliably measurable, but it obviously does
reduce the operation count in the normal (no exception
raised) path through PyEval_EvalFrameEx().
The tim-exc_sanity branch tries to push this harder, but
is still blowing up (at least in part due to pre-existing
subtle bugs that appear to have no other visible
consequences!).
Not a bugfix candidate.
invalid file paths for the built-in import machinery which leads to
fewer open calls on startup.
Also fix issue with PEP 302 style import hooks which lead to more open()
calls than necessary.
both mystrtoul.c and longobject.c. Share the table instead. Also
cut its size by 64 entries (they had been used for an inscrutable
trick originally, but the code no longer tries to use that trick).
In rare cases of strings specifying true values near sys.maxint,
and oddball bases (not decimal or a power of 2), int(string, base)
could deliver insane answers. This repairs all such problems, and
also speeds string->int significantly. On my box, here are %
speedups for decimal strings of various lengths:
length speedup
------ -------
1 12.4%
2 15.7%
3 20.6%
4 28.1%
5 33.2%
6 37.5%
7 41.9%
8 46.3%
9 51.2%
10 19.5%
11 19.9%
12 23.9%
13 23.7%
14 23.3%
15 24.9%
16 25.3%
17 28.3%
18 27.9%
19 35.7%
Note that the difference between 9 and 10 is the difference between
short and long Python ints on a 32-bit box. The patch doesn't
actually do anything to speed conversion to long: the speedup is
due to detecting "unsigned long" overflow more quickly.
This is a bugfix candidate, but it's a non-trivial patch and it
would be painful to separate the "bug fix" from the "speed up" parts.
discussion.
There are two places of documentation that still mention __context__:
Doc/lib/libstdtypes.tex -- I wasn't quite sure how to rewrite that without
spending a whole lot of time thinking about it; and whatsnew, which Andrew
usually likes to change himself.
- Warn-raise ImportWarning when importing would have picked up a directory
as package, if only it'd had an __init__.py. This swaps two tests (for
case-ness and __init__-ness), but case-test is not really more expensive,
and it's not in a speed-critical section.
- Test for the new warning by importing a common non-package directory on
sys.path: site-packages
- In regrtest.py, silence warnings generated by the build-environment
because Modules/ (which is added to sys.path for Setup-created modules)
has 'zlib' and '_ctypes' directories without __init__.py's.
MAXPATHLEN-sized buffers for various output-buffers (like to realpath()),
and that's correct on BSD platforms, but not Linux (which uses PATH_MAX, and
does not define MAXPATHLEN.) Cursory googling suggests Linux is following a
newer standard than BSD, but in cases like this, who knows. Using the
greater of PATH_MAX and 1024 as a fallback for MAXPATHLEN seems to be the
most portable solution.
exceptions that can't be raised any further, because (for instance) they
occur in __del__ methods. The coroutine tests in test_generators was
triggering this leak. Remove the leakers' testcase, and add a simpler
testcase that explicitly tests this leak to test_generators.
test_generators now no longer leaks at all, on my machine. This fix may also
solve other leaks, but my full refleakhunting run is still busy, so who
knows?
using a custom, nearly-identical macro. This probably changes how some of
these functions are compiled, which may result in fractionally slower (or
faster) execution. Considering the nature of traversal, visiting much of the
address space in unpredictable patterns, I'd argue the code readability and
maintainability is well worth it ;P
passing a string. Martin already fixed the actual crash by ensuring
Py_UNICODE is unsigned. As discussed on python-dev, this fix
removes the possibility of creating a unicode string from a raw buffer.
There is an outstanding question of how to fix the crash in 2.4.
that are suspended outside of any try/except/finally blocks to be
garbage collected even if they are part of a cycle. Generators that
suspend inside of an active try/except or try/finally block (including
those created by a ``with`` statement) are still not GC-able if they
are part of a cycle, however.
Re-enable all the tests in test_trace.py except one. Still not sure that these tests test what they used to test, but they pass. One failing test seems to be caused by undocumented line number table behavior in Python 2.4.
tracing/line number table in except blocks.
Reflow long lines introduced by col_offset changes. Update test_ast
to handle new fields in excepthandler.
As note in Python.asdl says, we might want to rethink how attributes
are handled. Perhaps they should be the same as other fields, with
the primary difference being how they are defined for all types within
a sum.
Also fix asdl_c so that constructors with int fields don't fail when
passed a zero value.
Explicitly clear all elements from arena->a_objects and remove
assert() that refcount is 1. It's possible for a program to get a
reference to the list via sys.getobjects() or via gc functions.
def foo((x)): was getting recognized as requiring tuple unpacking
which is not correct.
Add tests for this case and the proper way to unpack a tuple of one:
def foo((x,)):
test_inpsect was incorrect before. I'm not sure why it was passing,
but that has been corrected with a test for both functions above.
This means the test (and therefore inspect.getargspec()) are broken in 2.4.
objimpl.h, pymem.h: Stop mapping PyMem_{Del, DEL} and PyMem_{Free, FREE}
to PyObject_{Free, FREE} in a release build. They're aliases for the
system free() now.
_subprocess.c/sp_handle_dealloc(): Since the memory was originally
obtained via PyObject_NEW, it must be released via PyObject_FREE (or
_DEL).
pythonrun.c, tokenizer.c, parsermodule.c: I lost count of the number of
PyObject vs PyMem mismatches in these -- it's like the specific
function called at each site was picked at random, sometimes even with
memory obtained via PyMem getting released via PyObject. Changed most
to use PyObject uniformly, since the blobs allocated are predictably
small in most cases, and obmalloc is generally faster than system
mallocs then.
If extension modules in real life prove as sloppy as Python's front
end, we'll have to revert the objimpl.h + pymem.h part of this patch.
Note that no problems will show up in a debug build (all calls still go
thru obmalloc then). Problems will show up only in a release build, most
likely segfaults.
This will hopefully get rid of some Coverity warnings, be a hint to
developers, and be marginally faster.
Some asserts were added when the type is currently known, but depends
on values from another function.
PyCodec_IncrementalDecoder().
Factor out common code from PyCodec_Encoder()/PyCodec_Decoder(),
PyCodec_IncrementalEncoder()/PyCodec_IncrementalDecoder() and
PyCodec_StreamReader()/PyCodec_StreamWriter().
of tuple) that provides incremental decoders and encoders (a way to use
stateful codecs without the stream API). Functions
codecs.getincrementaldecoder() and codecs.getincrementalencoder() have
been added.
Neal Norwitz
Armin Rigo
Georg Brandl
Andrew M. Kuchling
Martin v. Löwis
Jeremy Hylton
Brett Cannon
Walter Dörwald
Tim Peters
Thomas Heller
Phillip J. Eby
Andrew MacIntyre
Neil Schemenauer
Hye-Shik Chang
Richard Jones
Fredrik Lundh
Thomas Wouters
Guido van Rossum
Skip Montanaro
Ronald Oussoren
Anthony Baxter
Nick Coghlan