Re: [PATCH] daemon: restore getpeername(0,...) use

From: Jeff King <peff@peff.net>
Date: Sep 10, 2012, 17:58 UTC
Message-ID: <20120910175846.GA16791@sigill.intra.peff.net>
In-Reply-To: <k2l7s5$gl9$1@ger.gmane.org>

On Mon, Sep 10, 2012 at 07:26:26PM +0200, Joachim Schmitz wrote:

Show 17 quoted lines

> >>as are non-const array intializers, e.g.:
> >>
> >>               const char *args[] = { editor, path, NULL };
> >>                                              ^
> >>".../git/editor.c", line 39: error(122): expression must have a
> >>constant value
> >>
> >>So git source is not plain C89 code (anymore?)
> >
> >I remember we excised a whole bunch of non-constant initializers at
> >some point because somebody's compiler was complaining. But I suppose
> >this one has slipped back in, because non-constant initializers are
> >so damn useful. And nobody has complained, which I imagine means
> >nobody has bothered building lately on those older systems that
> >complained.
> 
> OK, record my complaint then ;-)

Oops, did I say "complained"? I meant "sent patches". Hint, hint. :)

> At least some older release of HP NonStop only have C89 and are still in use
> 
> And tying to compile in plain C89 mode revealed several other
> problems too (e.g. size_t seems not to be typedef'd?)

I think it is a mistake to set -std=c89 (or whatever similar option your compiler supports). Like I said, we are not interested in being strictly C89-compliant. We are interested in working on real-world systems.

If your compiler complains in the default mode (or when it is given some reasonable practical settings), then that's something worth fixing. But if your compiler is perfectly capable of compiling git, but you choose to cripple it by telling it to be pedantic about a standard, then that is not git's problem at all.

-Peff