recent/new Plasma5 crashes under investigation

[Rex Dieter]

FYI,

KDE SIG is investigating some recent/serious reported crasher bugs in 
rawhide, notably:

"Could not sync environment to dbus." (startkde)
https://bugzilla.redhat.com/show_bug.cgi?id=1191171

and (an older one, but the situation has gotten worse recently):

Qt5 application crashes when connecting/disconnecting displays
https://bugzilla.redhat.com/show_bug.cgi?id=1083664
https://bugreports.qt.io/browse/QTBUG-44388


Initial theories are that gcc5 landing may have something to do with it, 
since the recent crashes go away reverting to pre-gcc5 builds of the same 
packages.


More debugging and testing is ongoing, and will post more when this is 
better understood.


-- Rex

[Sandro Mani]

On 11.02.2015 16:04, Rex Dieter wrote:
>  FYI,
>
>  KDE SIG is investigating some recent/serious reported crasher bugs in
>  rawhide, notably:
>
>  "Could not sync environment to dbus." (startkde)
>  https://bugzilla.redhat.com/show_bug.cgi?id=1191171
>
>  and (an older one, but the situation has gotten worse recently):
>
>  Qt5 application crashes when connecting/disconnecting displays
>  https://bugzilla.redhat.com/show_bug.cgi?id=1083664
>  https://bugreports.qt.io/browse/QTBUG-44388
>
>
>  Initial theories are that gcc5 landing may have something to do with it,
>  since the recent crashes go away reverting to pre-gcc5 builds of the same
>  packages. 
While gcc 5.0 exposed some issues in the Qt codebase which are actual 
bugs (reported in the above linked QTBUG-44388), there is also 
definitely a compiler bug involved here (one hint of this being that I'm 
able to compile c++ code with a derived class method calling a private 
function of a base class). I'm currently trying to narrow down the 
issue. In any event, this bug is causing most Qt5 applications to crash 
right on start (regardless of the display configuration).

Sandro

[Rex Dieter]

Rex Dieter wrote:
 
>  KDE SIG is investigating some recent/serious reported crasher bugs
> in
>  rawhide, notably:
>  
>  "Could not sync environment to dbus." (startkde)
>  https://bugzilla.redhat.com/show_bug.cgi?id=1191171
>  
>  and (an older one, but the situation has gotten worse recently):
>  
>  Qt5 application crashes when connecting/disconnecting displays
>  https://bugzilla.redhat.com/show_bug.cgi?id=1083664
>  https://bugreports.qt.io/browse/QTBUG-44388
>  
>  
>  Initial theories are that gcc5 landing may have something to do with it,
>  since the recent crashes go away reverting to pre-gcc5 builds of the same
>  packages. 

and now Qt(4) doesn't build either, bug,
https://bugzilla.redhat.com/show_bug.cgi?id=1192464

Any help, advice would be appreciated (particularly input from gcc 
maintainers).


FESCo, work on the Plasma5 change/feature is suffering due to this.

-- Rex

[Sandro Mani]

On 13.02.2015 15:53, Rex Dieter wrote:
>  Rex Dieter wrote:
>    
> > KDE SIG is investigating some recent/serious reported crasher bugs in
> > rawhide, notably:
> >
> > "Could not sync environment to dbus." (startkde)
> > https://bugzilla.redhat.com/show_bug.cgi?id=1191171
> >
> > and (an older one, but the situation has gotten worse recently):
> >
> > Qt5 application crashes when connecting/disconnecting displays
> > https://bugzilla.redhat.com/show_bug.cgi?id=1083664
> > https://bugreports.qt.io/browse/QTBUG-44388
> >
> >
> > Initial theories are that gcc5 landing may have something to do with it,
> > since the recent crashes go away reverting to pre-gcc5 builds of the same
> > packages.
>
>  and now Qt(4) doesn't build either, bug,
>  https://bugzilla.redhat.com/show_bug.cgi?id=1192464
>
>  Any help, advice would be appreciated (particularly input from gcc
>  maintainers).
>
>
>  FESCo, work on the Plasma5 change/feature is suffering due to this. 
I'm close
to having a small test case for the Qt5 crash - hope I can 
file a bug by this evening/tomorrow.

[Sandro Mani]

On 13.02.2015 16:04, Sandro Mani wrote:
>  On 13.02.2015 15:53, Rex Dieter wrote:
> > Rex Dieter wrote:
> >> KDE SIG is investigating some recent/serious reported crasher bugs in
> >> rawhide, notably:
> >>
> >> "Could not sync environment to dbus." (startkde)
> >> https://bugzilla.redhat.com/show_bug.cgi?id=1191171
> >>
> >> and (an older one, but the situation has gotten worse recently):
> >>
> >> Qt5 application crashes when connecting/disconnecting displays
> >> https://bugzilla.redhat.com/show_bug.cgi?id=1083664
> >> https://bugreports.qt.io/browse/QTBUG-44388
> >>
> >>
> >> Initial theories are that gcc5 landing may have something to do with 
> >> it,
> >> since the recent crashes go away reverting to pre-gcc5 builds of the 
> >> same
> >> packages.
> >
> > and now Qt(4) doesn't build either, bug,
> > https://bugzilla.redhat.com/show_bug.cgi?id=1192464
> >
> > Any help, advice would be appreciated (particularly input from gcc
> > maintainers).
> >
> >
> > FESCo, work on the Plasma5 change/feature is suffering due to this.
>  I'm close to having a small test case for the Qt5 crash - hope I can 
>  file a bug by this evening/tomorrow 
Gcc bug filed with test case: 
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65057 . In short, this 
appears to be a linker issue when building the Qt shared libraries with 
-fuse-ld=gold -shared -Wl,--dynamic-list

[Sandro Mani]

On 13.02.2015 22:21, Sandro Mani wrote:
>  On 13.02.2015 16:04, Sandro Mani wrote:
> >
> > On 13.02.2015 15:53, Rex Dieter wrote:
> >> Rex Dieter wrote:
> >>> KDE SIG is investigating some recent/serious reported crasher bugs in
> >>> rawhide, notably:
> >>>
> >>> "Could not sync environment to dbus." (startkde)
> >>> https://bugzilla.redhat.com/show_bug.cgi?id=1191171
> >>>
> >>> and (an older one, but the situation has gotten worse recently):
> >>>
> >>> Qt5 application crashes when connecting/disconnecting displays
> >>> https://bugzilla.redhat.com/show_bug.cgi?id=1083664
> >>> https://bugreports.qt.io/browse/QTBUG-44388
> >>>
> >>>
> >>> Initial theories are that gcc5 landing may have something to do 
> >>> with it,
> >>> since the recent crashes go away reverting to pre-gcc5 builds of 
> >>> the same
> >>> packages.
> >>
> >> and now Qt(4) doesn't build either, bug,
> >> https://bugzilla.redhat.com/show_bug.cgi?id=1192464
> >>
> >> Any help, advice would be appreciated (particularly input from gcc
> >> maintainers).
> >>
> >>
> >> FESCo, work on the Plasma5 change/feature is suffering due to this.
> > I'm close to having a small test case for the Qt5 crash - hope I can 
> > file a bug by this evening/tomorrow
>
>  Gcc bug filed with test case: 
>  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65057 . In short, this 
>  appears to be a linker issue when building the Qt shared libraries 
>  with -fuse-ld=gold -shared -Wl,--dynamic-list
>  
So for Qt5: passing -no-use-gold-linker to configure seems to be the way 
to go for the moment.

[Rex Dieter]

Sandro Mani wrote:

> >> I'm close to having a small test case for the Qt5 crash -
> hope I can
> >> file a bug by this evening/tomorrow
> >
> > Gcc bug filed with test case:
> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65057 . In short, this
> > appears to be a linker issue when building the Qt shared libraries
> > with -fuse-ld=gold -shared -Wl,--dynamic-list
> >
>  So for Qt5: passing -no-use-gold-linker to configure seems to be the way
>  to go for the moment. 

OK, new qt5-qtbase builds on the way:

%changelog
* Mon Feb 16 2015 Rex Dieter <rdieter(a)fedoraproject.org&gt; 5.4.0-13
- -no-use-gold-linker (f22+, #1193044)


as well as qt-4.8 gcc5-related build fixes:
%changelog
* Mon Feb 16 2015 Rex Dieter <rdieter(a)fedoraproject.org&gt; 1:4.8.6-24
- more gcc5 detection fixes, in particular, ensure same QT_BUILD_KEY as gcc4 
for now

Per 4.8, I'll be posting patches upstream for review, will assume that qt's 
default QT_BUILD_KEY will include g++-5 , but we won't change our f23 
packaging/patches until they agree/standardize on that.  (f22 builds will 
remain gcc4 compatible)

-- Rex

[Miloslav Trmač]

>  and now Qt(4) doesn't build either, bug,
>  https://bugzilla.redhat.com/show_bug.cgi?id=1192464
>  
>  Any help, advice would be appreciated (particularly input from gcc
>  maintainers).
>  
>  
>  FESCo, work on the Plasma5 change/feature is suffering due to this. 
(Is this still applicable after the -no-use-gold-linker workaround?)

Suffering in what way?  Likely to have KDE 5 not perfect but still good enough to ship? 
Likely to have to revert to KDE 4?  Likely to unable use KDE 5, and unable to revert
because Qt 4 can not be rebuilt?
    Mirek

[Rex Dieter]

Miloslav Trmač wrote:

> > and now Qt(4) doesn't build either, bug,
> > https://bugzilla.redhat.com/show_bug.cgi?id=1192464
> > 
> > Any help, advice would be appreciated (particularly input from gcc
> > maintainers).
> > 
> > 
> > FESCo, work on the Plasma5 change/feature is suffering due to this.
>  
>  (Is this still applicable after the -no-use-gold-linker workaround?) 
Yes, there is at least one unexplained critical issue remaining,

"Could not sync environment to dbus." (startkde)
https://bugzilla.redhat.com/show_bug.cgi?id=1191171

>  Suffering in what way?  Likely to have KDE 5 not perfect but still
> good
>  enough to ship?  
Suffering as in we've lost over a week (and counting) of effort spent 
chasing problems related to the introduction of gcc5 rather than improving 
and polishing Plasma instead.

-- Rex

[Rex Dieter]

Rex Dieter wrote:

>  Miloslav Trmač wrote:
>  
> >> and now Qt(4) doesn't build either, bug,
> >> https://bugzilla.redhat.com/show_bug.cgi?id=1192464
> >> 
> >> Any help, advice would be appreciated (particularly input from gcc
> >> maintainers).
> >> 
> >> 
> >> FESCo, work on the Plasma5 change/feature is suffering due to this.
> > 
> > (Is this still applicable after the -no-use-gold-linker workaround?)
>  
>  Yes, there is at least one unexplained critical issue remaining,
>  
>  "Could not sync environment to dbus." (startkde)
>  https://bugzilla.redhat.com/show_bug.cgi?id=1191171 
And this one is now reportedly fixed too, was apparently another symptom of 
the gold linker bug.

-- Rex