Re: [PATCH] libiberty: Ensure same `errno` is passed to `libiberty_vprintf_buffer_size()` and `vsprintf()`

[LIU Hao]

在 2026-2-9 21:22, Richard Biener 写道:

On Mon, Feb 9, 2026 at 1:39 PM LIU Hao <lh_mo...@126.com> wrote:

The MSVCRT `strtoul()` function, which is called indirectly by
`libiberty_vprintf_buffer_size()`, resets `errno` to zero upon success.

Isn't the better fix to save/restore around the strtoul() call then?  And best
have a comment explaining why we do it.

Sure, that's acceptable too. I'll send a new patch.

I had the idea that errno is never altered on success, but now I see
the strtoul manpage explicitly documents that.  I'm confused.

[1] https://pubs.opengroup.org/onlinepubs/9799919799/functions/strtoul.html



--
Best regards,
LIU Hao

POSIX says 'these functions shall not change the setting of errno if successful.' [1] But despite that, if an empty string is passed, it may still set errno to EINVAL.

OpenPGP_signature.asc
Description: OpenPGP digital signature