Bug 1138372

Created attachment 935285 [details]
reproducer

*** Bug 1139550 has been marked as a duplicate of this bug. ***

$ git log --reverse --oneline b0ea45863df9a89588cb99a27332a5f44178e861^.. \
    -- gtk/gtkgrid.c \
  | cat -n
     1  b0ea458 grid: Fix access to grid->priv member
     2  5e1a06d GtkGrid: Tweak size allocation
     3  cc86a7b Add gtk_grid_remove_{row,column}
     4  627685e GtkGrid: Support baseline alignment in GtkGrid
     5  ebbc779 GtkGrid: Add missing Since docs
     6  06c8e8f docs: fix some typos in newly introduced methods
     7  13858fd GtkGrid: Undo a size allocation tweak
     8  0899ef7 gtk: Use new macros for defining private data

I think there has been an attempt to fix this bug in upstream. See commit #2 above. RHEL7 doesn't have commit #2; the last commit it has is commit #1.

However, commit #2 seems to have broken other things, hence it was reverted in commit #7. Fedora 20 is probably a snapshot / fork from between commit #2 and commit #7.

In other words, Fedora 20 probably has a temporary fix that turned out to be wrong in the longer term. Both RHEL-7 and upstream have no (long term) fix for this issue.

I extracted the gtkgrid.c file from both the F20 and RHEL7 SRPMs (@ dist-git commits 09cb1783 and e4a95081, respectively), and compared them against various versions of the file in the upstream tree.

$ git show b0ea458:gtk/gtkgrid.c | diff -u - ~/tmp/rhel7-gtkgrid.c
<nothing>

$ git show 0899ef7:gtk/gtkgrid.c | diff -u - ~/tmp/f20-gtkgrid.c 
<nothing>

So, I was wrong, probably; apparently it's not 5e1a06d but another commit in that range.

I was wrong, Giuseppe's analysis or hunch was right. I debugged into gtk3 and it returns the correct value for top-attach. It's the impenetrable argument marshalling hairball where the retval goes lost, ie. when the correct int-valued top-attach property gets turned into constant 0.

Confirming, I'm bisecting this between pygobject3 3.8.2 and 3.10.2. The former breaks, the latter works.

This is the fix that's missing from RHEL-7:

commit 466567373289e6f141709f08efa80ba588d3d64a
Author: Simon Feltman <sfeltman.org>
Date:   Tue Jul 2 18:06:01 2013 -0700

    Re-add support for passing GValue's by reference
    
    Fix special casing when marshaling from a Python held GValue
    to a GValue argument intended for a function call.
    The re-factoring of GValue marshaling in commit #9e47afe459df942d9f
    broke this by always making a copy of the GValue. This removed the
    ability to retrieve values with functions like
    gtk_style_context_get_style_property.
    
    https://bugzilla.gnome.org/show_bug.cgi?id=701058

Please backport ASAP. Thanks.

Created attachment 936981 [details]
manual backport of upstream fix (broken out patch)

I verified Laszio's reproducer myself and have a pygobject3 commit ready to go. We're just waiting on an ack from QA.

still pending for the pm_ack?  Can someone please pm_ack this?

thanks for your prompt response.  7.0 is affected as well by this problem but it has no impact on virt-manager (I've noticed the issue after the virt-manager rebase in 7.1) so fine for us if it is not cloned to z stream.

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://rhn.redhat.com/errata/RHBA-2015-0536.html