Re: definition for _attribute() in remote.c

["Philip Oakley" <philipoakley@iee.org>]

From: "Jeff King" <peff@peff.net>
> On Mon, Apr 25, 2016 at 10:02:38PM +0100, Philip Oakley wrote:
>
>> I'm looking at getting Git for Windows to compile via Visual Studio
>> (https://github.com/git-for-windows/git/pull/256).
>>
>> However the use of __attribute() in remote.c at L1662
>> (https://github.com/git-for-windows/git/blob/master/remote.c#L1662) has 
>> got
>> me confused in that I can't see how the regular definition of 
>> __attribute()
>> is #included in this case. A definition is given in
>> git\compat\regex\regex_internal.h but doesn't appear to be on remote.c's
>> include path.
>>
>> The line was introduced by 3a429d0 (remote.c: report specific errors from
>> branch_get_upstream, 2015-05-21) which appears to be later than the 
>> previous
>> MSVC testers had looked at.
>
> It should be handled in git-compat-util.h, which is included by cache.h,
> which is included by remote.c.
>
> There we have:
>
>  #ifndef __GNUC__
>  #ifndef __attribute__
>  #define __attribute__(x)
>  #endif
>  #endif
>
> which should make it a noop on compilers which don't know about it. Is
> VS (or another file) setting __GNUC__?

It's not the __attribute__ definition (a Gnu C ism), rather its the 
__attribute variant, which has a definition in regex_internal.h, and is used 
in the regex code. It's that one that's used in remote.c that I can't fathom 
(i.e. how it worked in normally)

regex_internal.h#L160-164
#ifdef __GNUC__
# define __attribute(arg) __attribute__ (arg)
#else
# define __attribute(arg)
#endif

thus when the compilation get to remote.c#L1662 it fails to find that 
definition.

Should that line use the gnu extension name?

--
Philip