Re: [PATCH v2] travis-ci: build and test Git on Windows

From: Jeff King <peff@peff.net>
To: Lars Schneider <larsxschneider@gmail.com>
Cc: git@vger.kernel.org, Johannes.Schindelin@gmx.de,
	gitster@pobox.com, sxlijin@gmail.com
Subject: Re: [PATCH v2] travis-ci: build and test Git on Windows
Date: Fri, 24 Mar 2017 10:59:40 -0400	[thread overview]
Message-ID: <20170324145940.xqjmwxphwmtx67ug@sigill.intra.peff.net> (raw)
In-Reply-To: <20170324113747.44991-1-larsxschneider@gmail.com>

On Fri, Mar 24, 2017 at 12:37:47PM +0100, Lars Schneider wrote:

> I think I addressed all issues from the v1 review (see interdiff below)
> with one exception. The script still uses bash instead of sh. Something
> about this does not work in sh:
>     --output >(sed "$(printf '1s/^\xef\xbb\xbf//')" >cat >&3)
> 
> Does anyone know how to make this sh compatible?

Process substitution is a bash-ism. Just looking at this snippet I would
ask why you aren't just using stdout, but from the whole diff it looks
like you're using --write-out and want to keep the streams separate.

The POSIX shell way would be to make your own named pipe rather than
anonymous one:

  mkfifo data
  sed ... <data &
  curl ... --output data
  rm data

But I don't think relying on bash is that big a deal, as long as the
script is clearly marked with the correct #!-line (which it is).

The ">cat" isn't doing AFAICT, though.

-Peff