Issue #5640: Fix _multibytecodec so that CJK codecs don't repeat
error replacement returned by codec error callbacks twice in
IncrementalEncoder and StreamWriter.
R. David Murray [Thu, 2 Apr 2009 04:50:03 +0000 (04:50 +0000)]
Clarify that datetime strftime does not produce leap seconds and datetime
strptime does not accept it in the strftime behavior section of the
datetime docs.
Closes issue 2568.
Porting by hand since svnmerge disclaims all knowledge of the trunk commit.
PyImport_AppendInittab() took a char * as a first argument even though that
string was stored beyond the life of the call. Changed the signature to be
const char * to help make this point.
test_logging was blindly clearing the warnings filter. This caused
PendingDeprecationWarnings to be spewed all over by unittest.failIf*(). Fix
moves over to using warnings.catch_warning to protect the warnings filter.
........
The cgitb module had imports in its functions. This can cause deadlock with the
import lock if called from within a thread that was triggered by an import.
Fix issue 2522. locale.format now checks that it is passed
exactly one pattern, which avoids mysterious errors where it
had seemed to fail to do localization.
........
Add import_function method to test.test_support, and modify a number of
tests that expect to be skipped if imports fail or functions don't
exist to use import_function and import_module. The ultimate goal is
to change regrtest to not skip automatically on ImportError. Checking
in now to make sure the buldbots don't show any errors on platforms
I can't direct test on.
........
r70775 | r.david.murray | 2009-03-30 19:05:48 -0400 (Mon, 30 Mar 2009) | 4 lines
Change more tests to use import_module for the modules that
should cause tests to be skipped. Also rename import_function
to the more descriptive get_attribute and add a docstring.
........
r70856 | r.david.murray | 2009-03-31 14:32:17 -0400 (Tue, 31 Mar 2009) | 7 lines
A few more test skips via import_module, and change import_module to
return the error message produced by importlib, so that if an import
in the package whose import is being wrapped is what failed the skip
message will contain the name of that module instead of the name of the
wrapped module. Also fixed formatting of some previous comments.
........
r70874 | r.david.murray | 2009-03-31 15:33:15 -0400 (Tue, 31 Mar 2009) | 5 lines
Improve test_support.import_module docstring, remove
deprecated flag from get_attribute since it isn't likely
to do anything useful.
........
r70876 | r.david.murray | 2009-03-31 15:49:15 -0400 (Tue, 31 Mar 2009) | 4 lines
Remove the regrtest check that turns any ImportError into a skipped test.
Hopefully all modules whose imports legitimately result in a skipped
test have been properly wrapped by the previous commits.
........
r70877 | r.david.murray | 2009-03-31 15:57:24 -0400 (Tue, 31 Mar 2009) | 2 lines
get_file() no longer leaks a FILE structure. If given a file descriptor, it calls dup() to get a new handle. Then both the FILE object and the fd can be closed.
This is important, because otherwise, the leaked FILE object will be closed on process exit, causing assertions on Windows, e.g. in the test_multiprocessing.py regression test.
Add is_declared_global() which distinguishes between implicit and
explicit global variables.
........
r70809 | jeremy.hylton | 2009-03-31 09:48:15 -0400 (Tue, 31 Mar 2009) | 14 lines
Global statements from one function leaked into parallel functions.
Re http://bugs.python.org/issue4315
The symbol table used the same name dictionaries to recursively
analyze each of its child blocks, even though the dictionaries are
modified during analysis. The fix is to create new temporary
dictionaries via the analyze_child_block(). The only information that
needs to propagate back up is the names of the free variables.
Add more comments and break out a helper function. This code doesn't
get any easier to understand when you only look at it once a year.
........
Jeremy Hylton [Tue, 31 Mar 2009 14:35:53 +0000 (14:35 +0000)]
Simplify the Request class. The basic components of the parsed
Request are now available as public attributes, e.g. full_url and
host. The accessor methods are deprecated. The implementation
replace the complicated __getattr__ machinery with a _parse() method.
The response from an HTTP request is now an HTTPResponse instance
instead of an addinfourl() wrapper instance. The wrapper provided
minimal extract functionality and was undocumented. The API of
addinfourl() was preserved, except for close hooks, by adding a few
methods and public attributes to the HTTPResponse class.
Minor update to OSX build-installer script, needed
to ensure that the build will succeed in a clean
checkout and with a non-default deployment target.
........
Georg Brandl [Tue, 31 Mar 2009 04:16:10 +0000 (04:16 +0000)]
Fix segfaults when running test_exceptions with coverage tracing, caused by wrongly defining Exception.__context__ as a T_OBJECT structmember which does not set the member to NULL on None assignment, and generally does not do type checks. This could be used to crash the interpreter by setting any object to __context__. The same applies to __cause__. Also document the PyException_* functions.
Fix issue #4865: add /Library/Python/2.7/site-packages to
sys.path on OSX, to make it easier to share (some) installed
packages between the system install and a user install.
........
Guido van Rossum [Mon, 30 Mar 2009 22:01:35 +0000 (22:01 +0000)]
Issue #5604: non-ASCII characters in module name passed to
imp.find_module() were converted to UTF-8 while the path is
converted to the default filesystem encoding, causing nonsense.
Thanks to Andrew Svetlov.
(This time to the right branch. Will block duplicate merge to 3.0.2.)
Brett Cannon [Mon, 30 Mar 2009 21:30:26 +0000 (21:30 +0000)]
Make sure time.strptime only accepts strings (and document the fact like
strftime). Already didn't accept bytes but make the check earlier. This also
lifts the limitation of requiring ASCII.
Brett Cannon [Mon, 30 Mar 2009 19:57:15 +0000 (19:57 +0000)]
Fix importlib.machinery.PathFinder.find_module() to essentially skip over None
entries in sys.path_importer_cache. While this differs from semantics in how
__import__ works, it prevents any implicit semantics from taking hold with
users.
* Set a custom icon on the Python installer DMG
* Remove last traces of "MacPython"
* Add options to build different flavors of the installer
(still defaulting to a 2-way universal build that
runs on OSX 10.3)
........
Fix issue where 'make altinstall' or 'make install' would install everything
that needs to be installed on OSX (depending on the configure flags)
........
* Updates installed dependencies to latest releaases (bzip, zlib, ...)
* Changes code for updating folder icons from Python code
that uses the deprecated Carbon module to a much simpler
Cocoa program in Objective-C
........
projects / python / log