[mythtv] QT Slots Corruption Issue on Shield Tube

[David Hampton]

On Thu, 2022-03-10 at 13:13 +1100, Mark Spieth wrote:
> On 10/03/2022 9:49 am, Peter Bennett wrote:
> > 
> > On 3/9/22 16:33, Mark Spieth wrote:
> > > On 10/03/2022 3:07 am, Peter Bennett wrote:
> > > > This is the bug report:
> > > > 
> > > > https://github.com/MythTV/mythtv/issues/368
> > > > 
> > > > This only happens on the user's Shield Tube, but has been
> > > > happening 
> > > > for eight months. During playback he crashes at random times.
> > > > 
> > > > The backtrace and debug prints he has supplied show that an
> > > > address 
> > > > is being corrupted in the Qt slot system. In the first
> > > > parameter, 
> > > > the first digit of the address is changed from 0xa to 0x2. It 
> > > > crashes in moc_mythsocket.cpp while trying to build the
> > > > parameter 
> > > > list for calling MythSocket::ReadStringListReal or 
> > > > MythSocket::ReadReal. The call originates in 
> > > > MythSocket::ReadStringList or MythSocket::Read which use the
> > > > slot 
> > > > mechanism to call the "Real" versions of those methods, which
> > > > may be 
> > > > in a different thread.
> > > > 
> > > > Does anybody have any ideas how to proceed? I cannot recreate
> > > > it 
> > > > myself as I do not have a Shield Tube. Also I do not understand
> > > > how 
> > > > QT internal address lists could get corrupted.
> > > 
> > > Is it fixed in QT6 or latest?
> > > 
> > > Is it a regression? Should we downgrade QT? If so to which
> > > version?
> > > 
> > > I can do a QT update if I know it will help.
> > > 
> > > Mark
> > > 
> > > 
> > > _______________________________________________
> > > 
> > We can try the latest QT and see if that helps. I don't know the 
> > status of getting QT6 to work with MythTV. What is involved in 
> > building with QT6 ? Do we just update the download location?
> 
> There will be a whole new set of patches. May be lucky and no patches
> are required.
> 
> Just tried with 5.15.3 but wouldn't download. Will spend some time on
> the weekend updating to at least 5.15.3 and we'll see how that goes.
> 
> The current one is 5.14.1
> 
> I've also had issues with 32bit on firetv max so have been using
> leanfront.
> 
> It sounded like the issue is known but there is no patch yet.
> 
> Maybe Ill try qt6 just for fun. I seem to remember some work was done
> to make that work.

There's a major issue with the Qt6 version that I haven't run to ground
yet. (Or even tried recently, as last time I spent days banging my head
against my desk.)  If you run an operation on the front end that causes
a process to be forked off, then the display stops updating.
Mythfrontend is still running, keystrokes are still accepted, and you
can back all the way out and quit mythfrontend, but nothing changes on
the screen.  I believe I could trigger it reliably by trying to
retrieve the metadata on a video.  (As I said, its been a while.)  It
also happened at other times, but I could always trace it back to a
process being executed in the background.  The problem has to do with
the MythMainWindow code hiding and later re-showing the window.  Its
not a QT problem, as it doesn't happen when I hack the code and force
mythfrontend to use the Qt painter.  It does happen with both the
OpenGL and Vulkan painters that are part of the MythTV code.  Also, for
some reason this only happens when the frontend and backend are on
different systems.  Makes it harder to debug on a single development
system.

I thought I had filed an issue about this, but I can't find anything on
github.  I'll try and find time in the next couple of days to write it
up.

Compiling with qt6 is easy. All you should need is to cherry-pick
commit 004bdfb747 into your tree and then add the --qmake argument to
your call to configure.  That commit hacks out all the code in mythtv
that references classes that were removed in qt6.  All the real qt6
related changes were committed to the main branch a long time ago. 
There is a devel/qt6 branch you could check out that differs from
master by that one commit, but its a six weeks out of date and needs to
be updated.  I should probably just pull a new branch as the current
one was created a year ago.  Its already had four merges commits that
brought in over 800 commits from the main branch.  Probably best to
start fresh.

David

P.S.  The qt6 based code runs fine, as long as you don't accidentally
fork a process.  I've been running it on my secondary frontend box for
a year now and I forget that box is running qt6 until I trip over this
bug.