From: Jeff King <peff@peff.net>
Date: Fri, 13 May 2016 19:59:11 GMT
Subject: Re: t3404 static check of bad SHA-1 failure
Message-ID: <20160513195911.GE9890@sigill.intra.peff.net>
In-Reply-To: <xmqqwpmx91mb.fsf@gitster.mtv.corp.google.com>

On Fri, May 13, 2016 at 12:52:44PM -0700, Junio C Hamano wrote:

> I _think_ "test -z" should succeed according to POSIX, because
> 
>  (1) it is not "test -z string" because it lacks string,
> 
>  (2) it is not any of the other "test -<option> thing" because -z,
>     and
> 
>  (3) the only thing it matches in the supported form of "test" is
>      "test <string>" that tests if the <string> is not the null
>      string, and "-z" indeed is not the null string.
> 
> For the same reason, "test -n" succeeds.

Yeah, I think you're right; POSIX is pretty clear that this falls under
case 3. So that means "test -z" quietly does what we want. But it means
that "test -n" does the _opposite_ of what we want.

And sadly,

  git grep 'test -n [^"]'

is not empty.

> But working around older/broken shells is easy and the resulting
> script it more readable, so let's take this.  It makes the resulting
> code easier to understand even when we know we run it under POSIX
> shell.

Yep. The POSIX-explanation of what is going on might be worth putting in
the commit message for the "-z" case (i.e., it should work, but the
"why" is subtle).

-Peff