From: Jeff King <peff@peff.net>
Date: Tue, 14 Jun 2016 21:50:19 GMT
Subject: Re: [PATCHv3] gpg-interface: check gpg signature creation status
Message-ID: <20160614215019.GB22334@sigill.intra.peff.net>
In-Reply-To: <xmqq60tbaba5.fsf@gitster.mtv.corp.google.com>

On Tue, Jun 14, 2016 at 11:13:54AM -0700, Junio C Hamano wrote:

> Michael J Gruber <git@drmicha.warpmail.net> writes:
> 
> >  	bottom = signature->len;
> > -	len = strbuf_read(signature, gpg.out, 1024);
> > +	strbuf_read(signature, gpg.out, 1024);
> > +	strbuf_read(&err, gpg.err, 0);
> 
> Hmmmm, isn't this asking for a deadlock?  When GPG spews more than
> what would fit in a pipe buffer to its standard error (hence gets
> blocked), its standard output may not complete, and the we would get
> stuck by attempting to read from gpg.out, failing to reach the other
> strbuf_read() that would unblock GPG by reading from gpg.err?

Yeah, it definitely is a deadlock. I think we'd need a select loop to
read into multiple strbufs at once (we can't use "struct async" because
that might happen in another process).

-Peff