From: Jeff King <peff@peff.net>
Date: Fri, 30 Nov 2012 20:24:35 GMT
Subject: Re: [PATCH 5/5] launch_editor: propagate SIGINT from editor to git
Message-ID: <20121130202435.GA7933@sigill.intra.peff.net>
In-Reply-To: <50A00116.8060604@kdbg.org>

On Sun, Nov 11, 2012 at 08:48:38PM +0100, Johannes Sixt wrote:

> Am 11.11.2012 17:57, schrieb Jeff King:
> > @@ -51,6 +51,8 @@ int launch_editor(const char *path, struct strbuf *buffer, const char *const *en
> >  		sigchain_push(SIGINT, SIG_IGN);
> >  		ret = finish_command(&p);
> >  		sigchain_pop(SIGINT);
> > +		if (WIFSIGNALED(ret) && WTERMSIG(ret) == SIGINT)
> > +			raise(SIGINT);
> 
> The return value of finish_command() is already a digested version of
> waitpid's status value. According to
> Documentation/technical/api-run-command.txt:
> 
> . If the program terminated due to a signal, then the return value is
> the signal number - 128, ...
> 
> the correct condition would be
> 
> 		if (ret == SIGINT - 128)

Yeah, that is the same thing as WTERMSIG (which uses "ret & 0x7f") for
the range of -127..-1. I do not mind changing it to match run-command's
stated output, but I am curious whether there are systems where WTERMSIG
is not defined in the same way, and the code would break.

-Peff