Boost C++ Libraries: Ticket #3189: Python integer/long conversion to C++ unsigned int/long is broken

attachment set

anderson.lizardo@… — Thu, 18 Jun 2009 13:04:11 GMT

attachment → boost_python_unsigned_converter_fix.patch

Patch for bug #3189 (against boost trunk)

attachment set

anderson.lizardo@… — Fri, 10 Jul 2009 20:14:01 GMT

attachment → boost_python_unsigned_converter_fix_v2.patch

Fix builtin converters for unsigned integers (v2)

anderson.lizardo@… — Fri, 10 Jul 2009 20:16:53 GMT

Attached a new version (v2) of the patch. Changes:

Added more comments
Changed exception from ValueError to OverflowError, to match Python policy on similar conversions.
Improved test cases in libs/python/test/test_builtin_converters.py.

attachment set

anderson.lizardo@… — Mon, 13 Jul 2009 13:22:44 GMT

attachment → boost_python_unsigned_converter_fix_no_ctypes.patch

Remove dependency on ctypes for tests

anderson.lizardo@… — Mon, 13 Jul 2009 13:27:30 GMT

Attached a new patch (boost_python_unsigned_converter_fix_no_ctypes.patch) to be applied on top of the already applied one.

This removes the dependency on ctypes added by previous patch. It seems ctypes is not available on all architectures/OSes, so instead add static methods to test_builtin_converters.cpp that return the sizes of the C++ types we need.

Also do a few renames to functions added to test_builtin_converters.py so that the line length remains sane.

status changed; resolution set

Ralf W. Grosse-Kunstleve — Wed, 15 Jul 2009 05:13:41 GMT

status new → closed
resolution → fixed

Patches applied as svn revisions 54919 and 54923. Anderson, thank you very much, especially for the comprehensive tests!

status changed; resolution deleted

Dave Abrahams — Fri, 17 Jul 2009 19:23:12 GMT

status closed → reopened
resolution fixed

Replying to rwgk:

Patches applied as svn revisions 54919 and 54923. Anderson, thank you very much, especially for the comprehensive tests!

Ralf, to me this looks like an incorrect bug report. numeric_cast<T> where T is an unsigned type is supposed to throw a bad_numeric_cast if the argument is negative. Did you verify that the supplied tests fail before applying the patch?

status changed; resolution set

Dave Abrahams — Fri, 17 Jul 2009 19:25:21 GMT

status reopened → closed
resolution → fixed

Oh, I guessthe problem is that PyLong_AsUnsignedLong returns an unsigned long, thus making the result of the ?: expression unsigned. Oops.