Just pushed a new fix [de2d72f2] for an old problem. Using the undo
*toolbar button* in the matrix sometimes didn't refresh. Especially
with the playback pointer in the segment.
Please test latest git. Here's the procedure:
- Create a big segment. 8 bars or so.
- Open the matrix on that segment.
- Make sure the playback position pointer is in the segment.
- Add 30 or so random notes so that you can see them all at once.
- Use the *toolbar* undo button (not the toolbar undo history drop down,
not Ctrl+Z and not the menu) to remove the notes one by one, paying
attention to whether they go away each time.
- Confirm that each note goes away each time the undo toolbar button is
pressed.
This is an ancient patch from David Faure from 2022 that fell through
the cracks.
Complete details are in this thread on the devel mailing list:
https://sourceforge.net/p/rosegarden/mailman/rosegarden-devel/thread/2982dd28-4a1e-3f41-c94b-11e231c31...@tedfelix.com/
Ted.
On 6/6/22 9:47 AM, David Faure wrote:
Debugging this a little bit further (I like tricky bugs) shows a number of
surprising facts.
1) One undo step sometimes calls Segment::notifyRemove more than once.
This is because BasicCommand::unexecute() calls copyFrom() which does
m_segment->erase(m_segment->findTime(m_modifiedEventsStart),
m_segment->findTime(m_modifiedEventsEnd));
As the comment in BasicCommand.cpp:302 says, surely this copyFrom() should be
done without firing notifications?
Note however that this isn't the actual problem. It's just rather confusing
while debugging ;)
But even if MatrixScene::handleEventRemoved() is called multiple times,
the update() in there has no effect after the first call.
The first call sets d->updateAll to true (in QGraphicsScene), the following
calls return immediately because updateAll is true. It gets set to false
whenever actual drawing happens (QGraphicsScene::drawItems, called by
QGraphicsView::paintEvent).
2) While debugging deep within Qt, I found out that the "dirty" region of the
graphicsview's viewport is only 4 pixels wide, because
Panned::showPositionPointer calls viewport()->update(oldRect) which is a 4
pixels wide rectangle. Normally the graphicsview is repainted via the scene's
update() mechanism (which stores a dirtyRegion in QGraphicsView itself). I
think the heart of the bug is a collision between the two update mechanisms.
The scene thinks it's going to trigger a repaint of the whole view, but the
view has stored only a tiny update region.
I think it can be fixed by using scene()->update() instead.
Patch attached, please test (I have of course reverted the commit that removes
the update() call from MatrixScene::handleEventRemoved).
But this makes me wonder, why is the "pointer" drawn in drawForeground, after
these complex manual updates, instead of making the pointer a graphics item
and letting QGraphicsScene/QGraphicsView take care of updating correctly for
that item?
_______________________________________________
Rosegarden-devel mailing list
rosegarden-de...@lists.sourceforge.net - use the link below to unsubscribe
https://lists.sourceforge.net/lists/listinfo/rosegarden-devel
_______________________________________________
Rosegarden-user mailing list
Rosegarden-user@lists.sourceforge.net - use the link below to unsubscribe
https://lists.sourceforge.net/lists/listinfo/rosegarden-user
