Collin Funk summarized:
> Hi Yuqi,
>
> Yuqi Guo <gu...@ios.ac.cn> writes:
>
> > Hello friends, do you have any idea about this?
>
> Sorry, I forgot to respond. I'm moving this to bug-gnulib@gnu.org. I
> don't know enough about docker but perhaps someone else can help.
>
> The original report [1]:
>
> > File:
> >
> > gnulib-tests/test-localeconv.c
> >
> > Code:
> >
> > ```
> >
> > struct lconv *l = localeconv ();
> >
> > ASSERT (l->frac_digits == CHAR_MAX);
> > ASSERT (l->p_cs_precedes == CHAR_MAX);
> > ASSERT (l->p_sign_posn == CHAR_MAX);
> > ASSERT (l->p_sep_by_space == CHAR_MAX);
> > ASSERT (l->n_cs_precedes == CHAR_MAX);
> > ASSERT (l->n_sign_posn == CHAR_MAX);
> > ASSERT (l->n_sep_by_space == CHAR_MAX);
> >
> > ```
> >
> > Problem:
> >
> > On RISCV, the `char` is unsigned. Thus, the `CHAR_MAX` is 255 instead of
> > 127 on x86. However, the localeconv values remain the same on RISCV as
> > x86 (the values are 127). This makes the> assertions fail on RISCV.
>
> I asked what system and Yuqi says [2]:
>
> > I use a RISCV QEMU emulator to compile and run the findutils package.
> > The emulator was packed and published on Docker Hub by someone several
> > months ago. You can get it by `docker pull
> > xfan1024/openeuler:23.09-riscv64` and then start a docker container to
> > easily use it. Note, you may need to have `qemu-riscv64-static`
> > installed on your host system as this emulator is driven by it.
> >
> > Inside the emulator, I find that:
> > (1) The system is openEuler-23.09
> >
> > (2) The libc is GNU libc (version 2.38), and the compiler is gcc (version
> > 12.3.1)
>
> The correct behavior is these values should be CHAR_MAX as described by
> POSIX [3]:
>
> The members with type char are non-negative numbers, any of which can
> be {CHAR_MAX} to indicate that the value is not available in the
> current locale.
>
> Collin
>
> [1] https://lists.gnu.org/archive/html/bug-findutils/2024-06/msg00036.html
> [2] https://lists.gnu.org/archive/html/bug-findutils/2024-06/msg00039.html
> [3] https://pubs.opengroup.org/onlinepubs/009695399/functions/localeconv.html
I ran a testdir of Gnulib's 'localeconv' module, which includes this test,
on a Linux/riscv64 system (Fedora in my case; the openEuler 24.03/riscv64 ISO
image does not boot for me). It succeeds. Also, I looked for known glibc
bugs regarding localeconv: None. Thus, it is very unlikely that it is a glibc
bug.
To me, it's more probable that the gcc that was used to compile the glibc of
openEuler/riscv64 and the gcc that you use to compile your program are
incompatible. Namely, that one of them assumes that 'char' is signed and the
other one assumes that 'char' is unsigned.
Bruno
