Re: [PATCH] libio: Update number of written bytes in dprintf implementation

[Florian Weimer <fweimer@redhat.com>]

* Carlos O'Donell:

> On 1/31/23 04:38, Florian Weimer wrote:
>> The __printf_buffer_flush_dprintf function needs to record that
>> the buffer has been written before reusing it.  Without this
>> accounting, dprintf always returns zero.
>> 
>> Fixes commit 8ece45e4f586abd212d1c02d74d38ef681a45600
>> ("libio: Convert __vdprintf_internal to buffers.
>
> Tested that regression test catches the error without the patch applied.
>
> Tested that the fix corrects the error and correctly adjusts written with
> the number of bytes that were just written in the lines before in the flush.
> For dprintf we must flush everything since there is no buffer. The code
> added looks correct to me and mirrors __wprintf_buffer_flush_to_file (which
> adjusts written by count), __printf_buffer_flush_to_file (which always accounts
> for the bytes), __printf_buffer_flush_obstack (always counted object out bytes),
> and __printf_buffer_flush_snprintf (again always counted).
>
> Tested without regression on x86_64 and i686.
>
> OK for 2.37, pelase commit ASAP since I consider this a release blocker bug
> fix since it was reported by downstream testing in Fedora as part of the
> Fedora 38 Alpha testing (using latest glibc development).
>
> Reviewed-by: Carlos O'Donell <carlos@redhat.com>
> Tested-by: Carlos O'Donell <carlos@redhat.com>

Thanks, pushed.

Florian