Exception-handling bugs

[David_Abrahams at Dragonsys dot com]

Hi,

I am a maintainer contributor to the STLport (www.stlport.org), and in
particular I maintain its exception-handling test suite. This suite is
designed to test the C++ standard library components (we use it on the
STLport, not the standard gcc library) for correct behavior with respect
to exceptions, but it has also been very reliable at uncovering
compiler/code-generation bugs. I have been running this test with
several different builds of gcc 2.95.2 recently. The following builds
(at least) seem to have problems:

Very serious problems, crashes almost immediately under any
configuration:
     powerpc-unknown-linux (powerpc linux hosted and targeted).**
     mumit khan's port of gcc to Win32 (x86 WinNT hosted and targeted)

Aborts under one particular configuration in a manner seemingly
consistent with exceptions being thrown during unwinding or violated
exception-specification:
     x86 linux hosted and targeted

By the way, just to prove that the test suite itself isn't at fault, it
works under any configuration and optimization level with the following:
     sparc solaris hosted and targeted
     VC++6
     Metrowerks CodeWarrior targeting 68K

The test suite is not small, so I doubt that I can post it in a single
mail message. If anyone is interested in trying it, please contact me
directly, as I don't always watch this list carefully.

Thanks,
Dave

** trying to compile with optimizations also reveals a problem in one
particular source file, wherein the assembler complains of an "invalid
relocation mode"