Re: CVS commit: pkgsrc/mk/check

[Jonathan Perkin <jperkin%joyent.com@localhost>]

* On 2015-11-27 at 13:00 GMT, Jonathan Perkin wrote:

> * On 2015-11-27 at 12:55 GMT, Thomas Klausner wrote:
> 
> > On Fri, Nov 27, 2015 at 12:45:04PM +0000, Jonathan Perkin wrote:
> > > Module Name:	pkgsrc
> > > Committed By:	jperkin
> > > Date:		Fri Nov 27 12:45:04 UTC 2015
> > > 
> > > Modified Files:
> > > 	pkgsrc/mk/check: check-interpreter.mk
> > > 
> > > Log Message:
> > > Perform interpreter sed(1) check with LC_ALL=C, avoids failures on newer
> > > Darwin releases with "sed: RE error: illegal byte sequence".
> > 
> > Shouldn't we set LC_ALL=C in the general pkgsrc environment?
> 
> We already set it in ALL_ENV, but ${RUN} doesn't use that - I can
> certainly test a bulk build with it enabled.

Bulk builds for this change are now complete.  The change I
implemented was this:

  https://github.com/joyent/pkgsrc/commit/db05002

The before and after reports are here:

  http://us-east.manta.joyent.com/pkgsrc/public/reports/upstream-bulktest/20151129.2102/meta/report.html
  http://us-east.manta.joyent.com/pkgsrc/public/reports/upstream-bulktest/20151127.1333/meta/report.html

In total 4 packages failed with the ALL_ENV change which previously
built fine.  Two of them were transient failures:

  http://us-east.manta.joyent.com/pkgsrc/public/reports/upstream-bulktest/20151127.1333/tcl-pdflib-4.0.3nb17/build.log
  http://us-east.manta.joyent.com/pkgsrc/public/reports/upstream-bulktest/20151127.1333/xmonad-contrib-0.11.3nb4/build.log

due to resource exhaustion and possible MAKE_JOBS failure.  Running
these builds again worked in both cases.

The other two failed builds were affected by the ALL_ENV change, in
both cases because we expand $HOME in a generated file:

  http://us-east.manta.joyent.com/pkgsrc/public/reports/upstream-bulktest/20151127.1333/majordomo-1.94.5nb20/install.log
  http://us-east.manta.joyent.com/pkgsrc/public/reports/upstream-bulktest/20151127.1333/ruby200-puppet-3.8.4/install.log

Arguably majordomo is wrong and should be using a package-specific
MAJORDOMO_HOME or similar instead.  Puppet is generating it internally
so would need to be patched.

As it stands I'd be uncomfortable committing this change, as SmartOS
bulk builds still mask quite a lot of packages.  It would be helpful
if NetBSD users could attempt it too so we get a full picture of what
the fallout would be, and perhaps aim for after 2015Q4.

Cheers,

-- 
Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com