PyErr_Format() and PyUnicode_FromFormat() are able to allocate the right buffer
size and to catch memory allocation failures.

UTF-8 is just fine for module names.

I cannot use %U: fullname is a bytes object, not an unicode object. %A format
cannot be used, it adds 'b' (bytes) prefix. So create cant_find_module()
function to decode the filename and raise the error message.

It was not completly implemented. Add a test.

Use PyUnicode_FSConverter to support surrogates in the full name.

Don't hardcode path size in make_filename().

This adds $ABIFLAGS to the static library, since it should match the shared
library name.  Also, include the abiflags in python-config --libs.

use encoding other than ASCII or UTF-8.  Source encoding is preserved
when reindented code is written to a file.

instead of utf-8.

the unicode flag, to decode the filename, instead of the filesystem encoding.
Use the same choice than the zipfile module.

fixed root directory issue on posixpath.relpath(). (Ported working fixes
from ntpath)

... instead of a char*. Encode the module path to the fileystem encoding
(for PyImport_ExecCodeModuleEx) instead of utf-8.

PyImport_AddModule() returns a borrowed reference. Don't display "import ... #
loaded from Zip ..." on error.

Patch by Sébastien Sablé.  This solves a test_mmap failure on AIX.

Patch from ActiveState.

Patch and tests by Neil Muller.

And not the locale encoding. With this commit, the test finally pass on Windows
with a non-ascii path :-)

encodable to utf-8, not to the file1 encoding.

Tools/gdb/libpython.py.