RE: svn commit: r1601496 - /subversion/trunk/tools/dist/dist.sh

[Bert Huijben <be...@qqmail.nl>]

> -----Original Message-----
> From: breser@apache.org [mailto:breser@apache.org]
> Sent: maandag 9 juni 2014 22:03
> To: commits@subversion.apache.org
> Subject: svn commit: r1601496 - /subversion/trunk/tools/dist/dist.sh
> 
> Author: breser
> Date: Mon Jun  9 20:02:31 2014
> New Revision: 1601496
> 
> URL: http://svn.apache.org/r1601496
> Log:
> Have dist.sh generate subversion.pot for the benefit of translators who need
> it in the source distribution.
> 
> * tools/dist/dist.sh: Generate subversion.pot by running po-update.sh.

This patch appears to have broken the nightly snapshot building job.

http://ci.apache.org/builders/svn-trunk-nightly/builds/42
http://ci.apache.org/builders/svn-trunk-nightly/builds/42/steps/shell/logs/stdio
[[
subprocess.CalledProcessError: Command '['/usr/home/buildslave27/slave27/svn-trunk-nightly/uploads/dist.sh', '-v', 'nightly', '-pr', 'trunk', '-r', '1601546', '-zip', '-nightly']' returned non-zero exit status 1
]]

	Bert 

Re: svn commit: r1601496 - /subversion/trunk/tools/dist/dist.sh

[Ben Reser <br...@apache.org>]

On 6/10/14, 2:12 AM, Bert Huijben wrote:
> This patch appears to have broken the nightly snapshot building job.
> 
> http://ci.apache.org/builders/svn-trunk-nightly/builds/42
> http://ci.apache.org/builders/svn-trunk-nightly/builds/42/steps/shell/logs/stdio
> [[
> subprocess.CalledProcessError: Command '['/usr/home/buildslave27/slave27/svn-trunk-nightly/uploads/dist.sh', '-v', 'nightly', '-pr', 'trunk', '-r', '1601546', '-zip', '-nightly']' returned non-zero exit status 1
> ]]

Yup, it assumes that you can run shell scripts in the export.  Which works fine
for the tarballs.  But breaks on the zip file because we set the native
eol-style to CRLF.  The shell ends up interpreting the CR as part of the
command and doesn't run properly.

I've fixed this in r1601708 and r1601714 by making dist.sh separately export
po-update.sh like we do for find_python.sh.  Another option would have been
changing the eol-style on po-update from native to LF.  But I suspect this
would have been reverted by some well meaning person in the future, breaking
this again.  I have no idea if shell scripts being exported as CRLF is useful.
 In particular, do sh implementations on Windows deal with CRLF scripts?  If
they don't then we probably should make this change from native to LF for all
our shell scripts.