[Sugar-devel] Bug 1240354 - SoaS live x86_64 20150706 does not login from live system user

[Jerry Vonau]

> On September 4, 2015 at 12:37 PM Martin Abente
> <martin.abente.lahaye at gmail.com> wrote:
> 
> 
> Hello Peter,
> 
> On Fri, Sep 4, 2015 at 5:50 AM, Peter Robinson <pbrobinson at gmail.com>
> wrote:
> 
> >
> > On 4 Sep 2015 00:22, "Martin Abente" <martin.abente.lahaye at gmail.com>
> > wrote:
> > >
> > >
> > >
> > > On Thu, Sep 3, 2015 at 6:43 PM, James Cameron <quozl at laptop.org>
> > > wrote:
> > >>
> > >> The errors complain of missing symbols;
> > >>
> > >> - sugar_event_controller_get_state
> > >>
> > >> - acme_volume_alsa_new
> > >>
> > >> Which were both added after 0.98.
> > >>
> > >> On my F18 test systems, the package sugar-toolkit-0.98 brings in a
> > >> _sugarext.so file in /usr/lib/python2.7/site-packages/sugar/ which
> > >> does not have these symbols.
> > >>
> > >> If this file were being used instead, then the outcome may be as
> > >> reported.
> > >>
> > >> Is the package present?
> > >
> > >
> > > No, is not present. Also, that does not explain why the same 106
> > packages work perfectly fine on fc23 when are built for fc22, while
> > they
> > fail when built for fc23.
> > >
> > > As you mentioned in a previous reply, the only difference in build
> > > time
> > I also noticed, looking at the logs, is
> > the -specs=/usr/lib/rpm/redhat/redhat-hardened-ld' flag. which is
> > passed to
> > all gir-* (gobject introspection) calls.
> > >
> >
> > Yes, it's a well advertised feature of F-23, but we're not the first
> > distro to enable it and unless there's dodgy stuff going on in sugar it
> > should cause no issues.
> >
> 

Added both bugs listed in bugzilla as blockers for Bug 1199775 -
(harden-failure)

> I rebuilt [1] the sugar-toolkit-gtk3 106 packages for fc23, adding
> "%undefine
> _hardened_build" [2,3] to the .spec file, and now Sugar works fine on
> fc23.
>

Comparing build.logs on F23 with/without hardened_build these stick out vs
F22:
-DGLIB_MIN_REQUIRED_VERSION=GLIB_VERSION_2_42
-DGLIB_MAX_ALLOWED_VERSION=GLIB_VERSION_2_44

Think those flags get pulled into the buildroot via gtk3-devel[1] with the
change introduced in July[2] , glib2 with F22 having 2.44 and F23 having
2.45
 
> I could not figure out why having _hardened_build enable breaks sugar,
> but
> it looks like disabling could be an option for now.

I have a theory that hardened_build enforces those flags while 2.45 is
installed in the buildroot but doesn't fail the building.

Jerry


1. https://github.com/GNOME/gtk/blob/master/configure.ac
2. http://osdir.com/ml/commits.gnome/2015-07/msg04431.html