Is Boost-Python tested?

Dear list,

    during the last two days i have been compiling my research code
(around 500k lines of c++) using clang (clang 2.9 as downloaded from the
svn 2 days ago).
My code relies hevily on boost python.In doing my compilation i first
compiled boost1.44 with clang (which went fine) and i later used it in
compiling my code.

Apart for minor issues the compilation went fine and the code runs
correctly through the benchamrk example, HOWEVER on exit i get a
segmentation fault.
The segmentation fault is also confirmed by valgrind.

Our code has been tested (and works correctly) using gcc, intel, sun and
msvc (this last one in windows), and provides a pretty clean valgrind
output (at least as clean as the boost python hello world example).
I am honestly a bit surprised of having such a segfault on our test
examples, and i was wandering if the boost_python stuff has been tested,
in particular with reference to the allocation of "internal_references".

please find attached the output of valgrind for the case of interest,
where i just load our library in python and i close the program.

for reference the gcc output (intel's is very similar) finishes with

==6879== in use at exit: 1,866,386 bytes in 3,436 blocks
==6879== total heap usage: 9,373 allocs, 5,937 frees, 4,125,837 bytes
==6879== LEAK SUMMARY:
==6879== definitely lost: 0 bytes in 0 blocks
==6879== indirectly lost: 0 bytes in 0 blocks
==6879== possibly lost: 368,951 bytes in 361 blocks
==6879== still reachable: 1,497,435 bytes in 3,075 blocks
==6879== suppressed: 0 bytes in 0 blocks

As a secondary thing, i wanted to comment that the compilation time is
not better than gcc. I compiled the clang myself, so i probably did
somethign wrong. In particular i attempted habilitating optimization,
but i probably skipped something to have optimal performance. Any
suggestion on how to make it faster?

thank you in advance for any suggestion you can provide


clang_valgrind.log.tar.gz (9.89 KB)

I know the develops of Boost.Python; I can assure you it's very well tested.
The Boost.Python list might be able to help you.

- --
Bryce Lelbach aka wash