Re: [lmi] LLVM libc++

[Greg Chicares]

On 10/5/22 16:39, Vadim Zeitlin wrote:
> On Wed, 5 Oct 2022 15:58:03 +0000 Greg Chicares <gchicares@sbcglobal.net> 
> wrote:
> 
> GC> On 10/4/22 16:12, Vadim Zeitlin wrote:
> GC> > On Tue, 4 Oct 2022 15:02:26 +0000 Greg Chicares 
> <gchicares@sbcglobal.net> wrote:
> GC> > 
> GC> > GC> I'm doing something equivalent to your patch, but I've run into two
> GC> > GC> linker problems.
> GC> 
> GC> Without even addressing those problems yet, I've found another, which
> GC> is triggered by llvm's unwind library, with gcc and libstdc++'s regex
> GC> in combination with ubsan:
> 
>  Sorry, but how is this related to llvm libunwind? I must be missing
> something, but I just don't see how does it affect this?

This problem is observed in the presence of llvm libunwind,
but not observed in its absence (i.e., with HP libunwind).
That's all I know.

> GC> We've seen something like this before:
> GC> 
> GC>   commit 5a4fcaed7309f5d4caee3443538ceb3edf0e119b
> GC>   Author: Gregory W. Chicares <gchicares@sbcglobal.net>
> GC>   Date:   2022-08-18T16:22:15+00:00
> GC>   [...]
> GC>   +# gcc-12 gives spurious warnings when <regex> is used. See:
> GC>   +#   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105562
> GC>   +# For lmi, this seems to affect only UBSAN builds.
> GC> 
> GC> Now, with llvm's unwinder, it affects unit tests as well,
> GC> and the resolution is the same: '-Wno-maybe-uninitialized',
> GC> only if target-is-affected AND gcc AND ubsan.
> 
>  Could we perhaps disable it just for <regex> contents, i.e. use the usual
> pragma push/ignored/pop around it?

I'll look into that. Thanks for the suggestion.