From: Jeff King <peff@peff.net>
Date: Thu, 29 Mar 2012 17:15:25 GMT
Subject: Re: [PATCH 2/2] git: continue alias lookup on EACCES errors
Message-ID: <20120329171525.GB12318@sigill.intra.peff.net>
In-Reply-To: <CAH6sp9Pw75x6YrmEyLmbsbvHrbs8r6xSp3YC2NP-jOed-zZ3+g@mail.gmail.com>

On Thu, Mar 29, 2012 at 01:16:47PM +0200, Frans Klaver wrote:

> > Yes, we can differentiate after the fact. Though I think it ends up
> > being almost the same code as just implementing execvp in the first
> > place.
> 
> It will, but doesn't stock execv*() also provide access to shell
> builtins? If that's the case then I wouldn't be bothered by the extra
> bit of code we need to understand what execvp has been doing. I think
> it would be sane to keep sane_execvp a wrapper instead of a
> reimplementation.

No, definitely not. Handling builtins is the responsibility of the
shell, not of execvp. It is responsible for falling back to "/bin/sh
$file" if execve returns ENOEXEC.

Anyway, I think the last round I posted is good enough. It is
approaching execvp in complexity, but it is still a little bit simpler.
And because it's on the error code path, if we are incompatible the
worst thing we can screw up is the error message, not the actual exec.

-Peff