testsuite_allocator.h vs anonymous namespace

[Paolo Carlini]

Paolo Carlini wrote:

> I also suspected that and, to be sure, I tweaked a bit 
> testsuite_allocator.h to avoid any possible issue of that kind. 
> However, strangely, a similar problem persists for 
> pool_allocator/allocate_chunk.cc, which only involves a static 
> *local*. Weird. I wonder whether something is actually very subtly 
> broken in the compiler in the area of statics inizialization... We 
> should probably try to prepare a minimal testcase, but currently the 
> only way I can reproduce is with the full, large, testsuite PCH (and 
> generic locale model)...

Turned out to be due to lameness in the library: in some facets we were 
using setlocale / new even when the current locale was already "C". 
Therefore I adopted consistently the strategy already adopted 
elsewehere, and reverted my recent tweak to testsuite_allocator.h.

Tested x86_64-linux, gnu and generic locale model.

Paolo.

////////////////
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: CL_smart_new
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20070524/5c06d0d9/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch_smart_new
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20070524/5c06d0d9/attachment-0001.ksh>