From: Jeff King <peff@peff.net>
Date: Tue, 09 Jun 2009 08:53:28 GMT
Subject: Re: [PATCH v3 1/3] Introduce die_errno() that appends strerror(errno) to die()
Message-ID: <20090609085328.GA25721@coredump.intra.peff.net>
In-Reply-To: <200906091022.31738.trast@student.ethz.ch>

On Tue, Jun 09, 2009 at 10:22:30AM +0200, Thomas Rast wrote:

> > Aren't you assuming that strerror(errno) has no '%'-signs here, which is
> > what kicked off the whole discussion?
> 
> True, of course.  Hrm.
> 
> So do we go back to v2 (for 1/3) and ask future callers to never use
> custom formats with die_errno, or should I write a version that
> doubles the % characters while tacking the error message onto the
> format?

Personally, I think it is fine to assume that the format and arguments
to die(), die_errno(), and to the die_routine() are all
printf-compatible. That is how it has always been until now, and I don't
see any callers who would want to change that.

This is all internal to git.  It's not as if we are talking about a
library being called by arbitrary code, and we want to leave doors open
for arbitrary clients. libgit2 would probably take a different approach
(though I imagine their general strategy is never to die(), but to
always return an error condition, anyway).

> [It's a pity that it seems impossible to add an extra argument to the
> va_args...]

Agreed, but I think we are out of luck short of variadic macros (which
are a C99-ism, and even then have some issues).

-Peff