Boost logo

Boost Testing :

From: David Abrahams (dave_at_[hidden])
Date: 2005-06-23 13:10:32

Douglas Gregor <doug.gregor_at_[hidden]> writes:

> On Jun 23, 2005, at 8:24 AM, David Abrahams wrote:
>> Until just now, I had forgotten that sometimes you really do need a
>> python that's built with the same compiler you're testing for -- I
>> guess that's something we'll need to build into BBv2.
> Can you expand on that "sometimes"?
> All of the Python failures on gcc-3.4.4-linux disappeared once I built
> Python with it; should I do the same for the other gcc-*-linux
> variants?

I don't remember the details, unfortunately. The Intel engineers made
some unfortunate choice that prevents ... I don't know ... maybe
exceptions, from working right?

Yeah, here it is:

  C++ exception-handling across a pair of icc-compiled shared
  libraries opened with dlopen() fails when the main application was
  compiled with gcc.

Oh, and they have a compiler flag that supposedly fixes it:
-cxxlib-gcc. But that was causing ICEs in earlier version of Intel.
Don't know the current status.

So, if there are ABI compatibility issues between 3.4.4 and the G++
that compiled your Python, you might have problems. That said, I've
never had any trouble using Boost.Python compiled with 2.95 or 3.x.x
with any GCC-built Python executable as long as I was using the right

Dave Abrahams
Boost Consulting

Boost-testing list run by mbergal at