After an evaluation, GNOME has moved from Bugzilla to GitLab. Learn more about GitLab.
No new issues can be reported in GNOME Bugzilla anymore.
To report an issue in a GNOME project, go to GNOME GitLab.
Do not go to GNOME Gitlab for: Bluefish, Doxygen, GnuCash, GStreamer, java-gnome, LDTP, NetworkManager, Tomboy.
Bug 710833 - strange locale-uninstalled rules
strange locale-uninstalled rules
Status: RESOLVED FIXED
Product: pitivi
Classification: Other
Component: General
Git
Other Linux
: Normal normal
: 0.93
Assigned To: Pitivi maintainers
Pitivi maintainers
Depends on:
Blocks:
 
 
Reported: 2013-10-24 19:55 UTC by Colin Walters
Modified: 2013-10-30 21:56 UTC
See Also:
GNOME target: ---
GNOME version: ---


Attachments
build: Delete locale-uninstalled rule (1.04 KB, patch)
2013-10-24 19:55 UTC, Colin Walters
none Details | Review
manifest: Add pitivi (899 bytes, patch)
2013-10-24 20:56 UTC, Colin Walters
none Details | Review
build: Don't attempt locale-uninstalled bits if DESTDIR is set (970 bytes, patch)
2013-10-30 17:45 UTC, Colin Walters
committed Details | Review

Description Colin Walters 2013-10-24 19:55:24 UTC
Trying to add pitivi to Continuous, I get:

make[2]: Entering directory `/ostbuild/source/pitivi/_build/po'
/bin/sh /ostbuild/source/pitivi/install-sh -d /ostbuild/results.././locale
mkdir: cannot create directory '/ostbuild/results..': Read-only file system
mkdir: cannot create directory '/ostbuild/results..': Read-only file system
make[2]: *** [install-data-yes] Error 1
make[2]: Leaving directory `/ostbuild/source/pitivi/_build/po'
make[1]: *** [locale-uninstalled] Error 2
make[1]: Leaving directory `/ostbuild/source/pitivi/_build'
make: *** [install-recursive] Error 1
ob: pid 4347 exited with code 2
Build of pitivi failed

This looks like it comes from the locale-uninstalled rule in the toplevel Makefile.am; I can't fathom what it's trying to do, and git blame just points back at the first time i18n was introduced.
Comment 1 Colin Walters 2013-10-24 19:55:43 UTC
Created attachment 258051 [details] [review]
build: Delete locale-uninstalled rule

This breaks the build in Continuous, and I don't know why it'd be
necessary.  Hopefully pitivi's authors use jhbuild or the like and
thus install the translations.
Comment 2 Colin Walters 2013-10-24 20:56:33 UTC
Created attachment 258057 [details] [review]
manifest: Add pitivi

More apps!
Comment 3 Colin Walters 2013-10-24 20:57:08 UTC
Attaching the manifest patch here just for my own reference, so once this bug is fixed it can get on the Continuous train.
Comment 4 Mathieu Duponchelle 2013-10-30 17:03:17 UTC
Disclaimer : I am not an autotools expert.

That patch still looks suspicious to me, is there a good reason for deleting that rule except "it makes stuff work for me" ?

We don't use jhbuild, and use pitivi completely uninstalled without any problem, and we would like to keep it that way.

Also, it's kind of hard for us to reproduce the issue, as "Trying to add pitivi to Continuous" is not really a way to reproduce, can you tell us what continuous is trying to do that fails, that way we'll be able to fix that bug ?
Comment 5 Colin Walters 2013-10-30 17:32:51 UTC
(In reply to comment #4)

> That patch still looks suspicious to me, is there a good reason for deleting
> that rule except "it makes stuff work for me" ?

See below.

> We don't use jhbuild, and use pitivi completely uninstalled without any
> problem, and we would like to keep it that way.

Ok.  I understand that, and I don't want to break it.  (Personally though I gave up on it for my apps, and I just use "jhbuild make" to install in the jhbuild root)

> Also, it's kind of hard for us to reproduce the issue, as "Trying to add pitivi
> to Continuous" is not really a way to reproduce, can you tell us what
> continuous is trying to do that fails, that way we'll be able to fix that bug ?

The executive summary is gnome-continuous is a package-like build system, except automated continuously.

What's very important is that it builds every module in *exactly* the same way:
env NOCONFIGURE=1 ./autogen.sh
./configure --prefix=/usr --sysconfdir=/etc
make
make install DESTDIR=/tmp/tmpdestdir

Now since pitivi is packaged, presumably this should fail in the same way inside dpkg/rpm...except now that I look, it doesn't.

From the latest koji log in Fedora:
http://kojipkgs.fedoraproject.org//packages/pitivi/0.15.2/5.fc21/data/logs/noarch/build.log

if test -d po; then \
    make -C po datadir=../. install; \
fi

appears to work.  Except if you look carefully, it's creating files outside of the buildroot:

installing ar.gmo as /builddir/build/BUILDROOT/pitivi-0.15.2-5.fc21.noarch.././locale/ar/LC_MESSAGES/pitivi.mo

Which Continuous is more strict about.

Apparently it works in Debian too, I don't see any patches for this in the sid package.

(They do carry a patch to disable pitivi's "devel mode")
Comment 6 Colin Walters 2013-10-30 17:45:46 UTC
Created attachment 258601 [details] [review]
build: Don't attempt locale-uninstalled bits if DESTDIR is set

How about this patch?  It's a lot simpler and should hopefully avoid
breaking you running uninstalled.
Comment 7 Mathieu Duponchelle 2013-10-30 19:44:27 UTC
This looks way less intrusive to me already, I don't see no reason why we shouldn't merge this now. As stated before, I'm far from autotools mastery, and none of the two other current maintainers are either, so there might be an issue here that I could fail to see, thanks for the patches anyway and sorry for the low reactivity :)
Comment 8 Colin Walters 2013-10-30 21:36:23 UTC
Thanks!  I'll take that as an accepted-commit-now even though you
didn't change the patch status in Splinter.  (Do you guys use
Splinter/Bugzilla for patches?)

Please ping me on IRC (walters) or reopen if you have any issues!

Attachment 258601 [details] pushed as 5c94a24 - build: Don't attempt locale-uninstalled bits if DESTDIR is set
Comment 9 Thibault Saunier 2013-10-30 21:41:22 UTC
(In reply to comment #8)
> Thanks!  I'll take that as an accepted-commit-now even though you
> didn't change the patch status in Splinter.  (Do you guys use
> Splinter/Bugzilla for patches?)

OK, I am OK with that, not sure it was the perfect patch, but it should not break our stuff, hopefully :)

about splitter/Bugzilla, it mainly depends on how big branches are, handling a big patch set in here is not exactly great so in that case we handle it differently, but in the case of a single patch, it is usually how we receive contributions yes.

Thanks for your patch.
Comment 10 Mathieu Duponchelle 2013-10-30 21:45:19 UTC
That was not an accepted-commit-now for the record, but as thibault doesn't seem to be against that patch either then no harm done.
Comment 11 Colin Walters 2013-10-30 21:51:12 UTC
So while pitivi builds in Continuous now, it doesn't actually run due to the dependencies on PyXDG and NumPy.  The first is really duplicative of what we already have in GLib, so I'll work on a patch to drop it unless someone objects.

Looks like NumPy is already a "soft" dependency.

Will open further bugs for any future work.
Comment 12 Thibault Saunier 2013-10-30 21:55:22 UTC
(In reply to comment #11)
> So while pitivi builds in Continuous now, it doesn't actually run due to the
> dependencies on PyXDG and NumPy.  The first is really duplicative of what we
> already have in GLib, so I'll work on a patch to drop it unless someone
> objects.

I would be happy to drop that dep :)
 
> Looks like NumPy is already a "soft" dependency.

It is a runtime dep, and right now we do not actually need it as it is for a feature that has not been reimplement yet after the port to GES.