Re: [RFC/PATCH] Add the NO_SENTINEL build variable

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 18, 2013, 21:25 UTC
Message-ID: <7vli531ryh.fsf@alter.siamese.dyndns.org>
In-Reply-To: <51E82582.1030204@ramsay1.demon.co.uk>

Ramsay Jones <ramsay@ramsay1.demon.co.uk> writes:

Show 24 quoted lines

> Jonathan Nieder wrote:
>> Ramsay Jones wrote:
>> 
>>> One of the three gcc compilers that I use does not understand the
>>> sentinel function attribute. (so, it spews 108 warning messages)
>> 
>> Do you know what version of gcc introduced the sentinel attribute?
>> Would it make sense for the ifdef in git-compat-util.h to be keyed on 
>> __GNUC__ and __GNUC_MINOR__ instead of a new makefile flag?
>> 
>
> I have on old (v4.2.1) gcc repo on Linux and looking at
>
>     ~/gcc-4.2.1/gcc/ChangeLog-2004
>
> I can see that the sentinel attribute was added on 2004-09-04 by
> Kaveh R. Ghazi.
>
> Also, I find "bump version string to version 4.0.0" was on 2004-09-09
> and "bump version string to version 3.5.0" was on 2004-01-16.
>
> Several of my system header files (on Linux) imply that the
> sentinel attribute is supported by __GNUC__ >= 4. (One of them,
> ansidecl.h, states that gcc 3.5 supports it but ...)

Perhaps a message from yesterday would have helped?

 http://article.gmane.org/gmane.comp.version-control.git/230633

seems to indicate that checking for version 4 is sufficient.

Also I asked you to split the __attribute__((sentinel(n)) support into a separate patch. We currently do not pass anything but 0 (meaning, the sentinel is always at the end), and SENTINEL in all capital is easy enough to grep for when somebody _does_ want to have such a support, so I'd prefer not to see __attribute__((sentinel(n)) until it becomes necessary.

Thanks.