Re: [PATCH] sparse: ignore warning from new glibc headers

From: Ramsay Jones <ramsay@ramsayjones.plus.com>
Date: Jul 19, 2024, 02:03 UTC
Message-ID: <28fd393a-2d8a-4019-95bc-d1a2e409cf53@ramsayjones.plus.com>
In-Reply-To: <xmqqplrb1f2p.fsf@gitster.g>

On 18/07/2024 05:47, Junio C Hamano wrote:

Show 13 quoted lines

> Ramsay Jones <ramsay@ramsayjones.plus.com> writes:
> 
>> On 17/07/2024 23:53, Junio C Hamano wrote:
>> [snip]
>>> That's OK.  So in short, with a separate SP_EXTRA_FLAGS with "-Wno-vla",
>>> Luc's patch is a sufficient fix without any downsides, no?
>>>
>>
>> Yes, assuming you're only concerned with 'make sparse' usage.
> 
> Is there anything else in the context of this project I should be
> concerned with, wrt sparse and recent </usr/include/regex.h> that
> uses vla in prototype parameters?

No, I don't think so. At the time I remember looking around for more uses of VLA's in the glibc header files and finding nothing of concern:

  $ find /usr/include -iname '*.h' |
  > xargs grep -n __STDC_NO_VLA__
  /usr/include/regex.h:527:	&& !defined __STDC_NO_VLA__)
  /usr/include/brotli/port.h:251:    !defined(__STDC_NO_VLA__) && !defined(__cplusplus) &&         \
  $

The <regex.h> header has only a single use of the _REGEX_NELTS macro in the declaration of the regexec() function.

The <brotli/port.h> header defines an BROTLI_ARRAY_PARAM macro, but does not use it in the header file. It is available for use in your own source file, but ...

I suppose it is possible that other headers use VLA's in their declarations and are not protected by __STDC_NO_VLA__, but that seems unlikely. The headers that we actually use (for the symbols we use), can't include any VLA's or gcc would complain (via -Wvla).

Also, it seems <regex.h> is the only header that suppresses -Wvla:

  $ find /usr/include -iname '*.h' |
  > xargs grep -n '\-Wvla'
  /usr/include/regex.h:536:# pragma GCC diagnostic ignored "-Wvla"
  $

ATB, Ramsay Jones