The build of xmms2 fails because of HASH(0x82db558)="" in the environment

On Wed, May 16, 2007 at 05:14:10PM +0200, Michael Bienia wrote:
> On 2007-05-16 12:55:43 +0100, Matt Zimmerman wrote:
> > If that's what it is, you'll probably need to look at config.log to find out
> > what's wrong; one way to debug this type of failure is to do something
> > analogous to:
> >
> > ./configure ... || (cat config.log; false)
>
> I've done it for xmms2 already which uses also scons to build and fails
> only on the Ubuntu buildds (but builds fine in a pbuilder and in
> Debian).
>
> The result was that the environment had a HASH(0x82db558)="" (the hex
> value differs between builds) which /bin/sh didn't like and errors out
> which scons interprets as not found.
>
> After consulting #ubuntu-devel I got told that this it probably a bug in
> the sbuild on the buildds and that I should ask Adam Conrad about it.
> I contacted him and told him about this problem and wait now till he
> finds some time to look at it and fix it.
> I've also already opened a bug about it:
> https://bugs.launchpad.net/launchpad-buildd/+bug/87077
>
> I assume that other packages using scons trigger this bug also and that
> at least aqsis and a package I already forgot have probably the same
> problem (I didn't upload a new version to get the config.log to confirm
> it).

The build log URL in that bug report isn't valid, and the one I could find:

http://librarian.launchpad.net/7540965/buildlog_ubuntu-gutsy-i386.xmms2_0.2DrHouse-3.1ubuntu3_FAILEDTOBUILD.txt.gz

doesn't show that line. I'd like to see the full build log.

That environment value looks like a bug in a Perl program. It shouldn't
break sbuild, but it shouldn't be there either.

Here's a copy of a failed build log, so that it is preserved (the ones in the librarian tend to vanish)

This problem is now reported also in scons upstream:

http://scons.tigris.org/issues/show_bug.cgi?id=1689

The attached patch should be sufficient to avoid the problem until it is fixed properly: it removes the use of os.system() entirely, causing SCons to use fork()+exec() on systems where spawnvpe() is not available, avoiding the need to escape anything.

Michael: can you check if the patch is ok and good to go?

Here is an updated debdiff. The changes are: fix syntax for bug closing and change the maintainer field.

I've tested it with a patched scons on my PPA and two packages (ardour and aqsis) which didn't build with the scons from the archive did build with the patched scons.

Daniel: can you sponsor the upload?

scons (0.97-1ubuntu1) gutsy; urgency=low

  [ Mark Brown ]
  * Don't use os.system()+env to spawn tools, working around underquoting
    of the environment until a real fix is done upstream (LP: #87077).

  [ Michael Bienia ]
  * debian/control: Modify Maintainer value to match
    DebianMaintainerField spec.

 -- Michael Bienia <email address hidden> Sat, 21 Jul 2007 01:20:19 +0200

This is fixed in launchpad-buildd version 40, which I've just rolled out to the entire buildd farm, primary and PPA.