Re: [PATCH] gdbsupport: bump unordered_dense library to 4.6.0

[Hannes Domani <ssbssa@yahoo.de>]

 I somehow missed Pedro's earlier reply.


Am Montag, 27. Oktober 2025 um 17:03:28 MEZ hat Simon Marchi <simon.marchi@polymtl.ca> Folgendes geschrieben:

> On 10/8/25 3:19 PM, Pedro Alves wrote:
> > Hi!
> >
> > On 2025-10-08 18:48, Hannes Domani wrote:
> >
> >>> +// <memory_resource> includes <mutex>, which fails to compile if
> >>> +// targeting GCC >= 13 with the (rewritten) win32 thread model, and
> >>> +// targeting Windows earlier than Vista (0x600).  GCC predefines
> >>> +// _REENTRANT when using the 'posix' model, and doesn't when using the
> >>> +// 'win32' model.
> >>> +#    if defined __MINGW64__ && defined __GNUC__ && __GNUC__ >= 13 && !defined _REENTRANT
> >>> +// _WIN32_WINNT is guaranteed to be defined here because of the
> >>> +// <cstdint> inclusion above.
> >>> +#        ifndef _WIN32_WINNT
> >>> +#            error "_WIN32_WINNT not defined"
> >>> +#        endif
> >>> +#        if _WIN32_WINNT < 0x600
> >>> +#            define ANKERL_MEMORY_RESOURCE_IS_BAD() 1 // NOLINT(cppcoreguidelines-macro-usage)
> >>> +#        endif
> >>> +#    endif
> >>> +#    ifndef ANKERL_MEMORY_RESOURCE_IS_BAD
> >>> +#        define ANKERL_MEMORY_RESOURCE_IS_BAD() 0 // NOLINT(cppcoreguidelines-macro-usage)
> >>> +#    endif
> >>
> >> This mentions '(rewritten) win32' and 'posix' thread models.
> >
> > The win32 thread model was completely rewritten in GCC 13, that's what I meant
> > by '(rewritten) win32'.  It seems like this isn't very common knowledge, which IMHO leads
> > to people still using other models when they probably don't need to.
> >
> > See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115105

Whoa, I subscribe to the gcc mailing lists, and read the gcc changes for new releases, but I didn't know that.
If the changes document was modified later, it explains at least why I missed it there.


> >> What about the 'mcf' thread model,
> >
> > No sure.  I've never used it.  Did that ever make it to GCC upstream?  I'm under the impression it did not.

Yes, it's in GCC upstream, since gcc-14 [1].
Also something that probably should've been in the gcc-14-changes document.


> > I don't really know if there is still a point in using that model vs stock win32 since GCC 13.

Now I'm a bit curious what the differences are, both support C++11 thread from Vista onward (but I already modified mcf so it works for XP as well).


> >> or is that meant with 'rewritten win32'?
> >
> > No.
> >
> > Pedro Alves
> >
> 
> I just remembered about this patch.  I was not sure if this discussion
> was going to go further, so I didn't push the patch right away at the
> time, but I just pushed it now.
> 
> In the mean time, there were some 4.7.0 and 4.8.0 releases, but I didn't
> see anything we needed in there.


[1] https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=f036d759ecee538555fa8c6b11963e4033732463

Hannes