[Linaro-TCWG-CI] gcc-15-8035-g7ee31bc9276: 2 regressions on master-thumb_m23_soft_eabi

[Jonathan Wakely]

On Mon, 17 Mar 2025 at 09:53, Jonathan Wakely <jwakely@redhat.com> wrote:
>
>
>
> On Monday, 17 March 2025, Christophe Lyon <christophe.lyon@linaro.org> wrote:
> > On Sun, 16 Mar 2025 at 21:54, Jonathan Wakely <jwakely@redhat.com> wrote:
> >>
> >> On Sun, 16 Mar 2025 at 13:16, <ci_notify@linaro.org> wrote:
> >> >
> >> > Dear contributor,
> >> >
> >> > Our automatic CI has detected problems related to your patch(es). Please find some details below.
> >> >
> >> > In  arm-eabi cortex-m23 soft, after:
> >> >   | commit gcc-15-8035-g7ee31bc9276
> >> >   | Author: Jonathan Wakely <jwakely@redhat.com>
> >> >   | Date:   Thu Mar 13 13:34:55 2025 +0000
> >> >   |
> >> >   |     libstdc++: Implement <stdbit.h> for C++26 (P3370R1)
> >> >   |
> >> >   |     This is the first part of the P3370R1 proposal just approved by the
> >> >   |     committee in Wrocław. This adds C++ equivalents of the functions added
> >> >   |     to C23 by WG14 N3022.
> >> >   | ... 16 lines of the commit log omitted.
> >> >
> >> > Produces 2 regressions:
> >> >   |
> >> >   | regressions.sum:
> >> >   | Running libstdc++:libstdc++-dg/conformance.exp ...
> >> >   | FAIL: 20_util/stdbit/1.cc -std=gnu++26 (test for excess errors)
> >> >   | UNRESOLVED: 20_util/stdbit/1.cc -std=gnu++26 compilation failed to produce executable
> >> >
> >> > Used configuration :
> >> >  *CI config* tcwg_gnu_embed_check_gcc arm-eabi -mthumb -march=armv8-m.base -mtune=cortex-m23 -mfloat-abi=soft -mfpu=auto
> >> >  *configure and test flags:* --target arm-eabi --disable-multilib --with-mode=thumb --with-cpu=cortex-m23 --with-float=soft --target_board=-mthumb/-march=armv8-m.base/-mtune=cortex-m23/-mfloat-abi=soft/-mfpu=auto qemu_cpu=cortex-m33
> >> >
> >> > We track this bug report under https://linaro.atlassian.net/browse/GNU-1543. Please let us know if you have a fix.
> >>
> >> All the errors are of the form:
> >> error: 'ULLONG_MAX' was not declared in this scope
> >> but the test includes <limits.h>.
> >>
> >> So this target doesn't support long long? Or just doesn't define ULLONG_MAX?
> >>
> >
> > It does...
> >
> > I've manually reproduced it, and ISTM the problem is __STDC_VERSION__
> > is not defined,
> > as gcc/glimits.h expects:
> > https://gcc.gnu.org/git/?p=gcc.git;a=blob;f=gcc/glimits.h;h=d5877602bf741383cfddb13236fbba1cf0b5b520;hb=HEAD#l102
>
>
> Aha! Thanks.
>
> > Compiling
> > ==============
> > #include <limits.h>
> > unsigned long long var = ULLONG_MAX;
> > ================
> > works with the same compiler, in C mode.
> >
> > But why would that work on arm-linux-gnueabihf and not on arm-none-eabi?
>
> I think glimits.h use only used if libc doesn't provide one, and I guess glibc's limits.h is used for gnueabihf
>
> The C++ standard says it's implementation-defined whether __STDC_VERSION__ is defined by a C++ compiler, and if defined, it's implementation-defined what is value is
>
> GCC/glimits.h should check || (defined(__cplusplus) && __cplusplus >= 201103L))
>
> i.e. long long should be supported for C++11 and later.
>
> Libstdc++ actually assumes long long is always supported even for C++98 so I'm surprised we've never noticed this before! I think we probably use the type without using the LLONG_MAX macro, so it just happens to work.
>
> I can adjust the test to be agnostic to that macro, but I'll also propose a patch to check __cplusplus in glimits.h
>
> Thanks again for finding the cause here.

Hmm, except that libstdc++ provides <climits> which should add
ULLONG_MAX if it's not defined by libc:
https://gcc.gnu.org/cgit/gcc/tree/libstdc++-v3/include/c_global/climits
And <limits.h> should find the libstdc++ version which includes
<climits>., but we're not installing the libstdc++ version of
<limits.h>.
That's a libstdc++ bug.