Bug 386726

Ooh, a video.

Can I tempt you to see if you can make a backtrace as well? That would point us to the exact line that crashes! https://docs.krita.org/Dr._Mingw_debugger

Created attachment 108779 [details]
The Krita log file from the %appdata% folder.

Sure thing, I didn't even know that a log file existed for this. Yes, that would truly be helpful.

Please note, that I did open Krita up a couple more times after I recorded the video, mostly to make sure that the settings were the way I wanted them to be the next time I wanted to use the program.

(In reply to wolthera from comment #1)
> Ooh, a video.
> 
> Can I tempt you to see if you can make a backtrace as well? That would point
> us to the exact line that crashes! https://docs.krita.org/Dr._Mingw_debugger


Sorry about the posting of the comment, I meant to have sent it as a reply. (This is my first time using this bug report system; I'm not too knowledgeable about it.)

I attached the log file, and I really hope it can be of help to you and the other developers.

Hm, that crash log isn't related to anything to do with strokes or smoothing; something weird happens when updating the display. I'm wondering what that might be, though. How do you exactly switch stabilizer mode when drawing? With a shortcut?

(In reply to Boudewijn Rempt from comment #4)
> Hm, that crash log isn't related to anything to do with strokes or
> smoothing; something weird happens when updating the display. I'm wondering
> what that might be, though. How do you exactly switch stabilizer mode when
> drawing? With a shortcut?

That is correct, I use the CTRL + (1,2,3,4) shortcuts to switch between smoothing engines. This problem does not happen if I stop drawing and manually switch the smoothing engine by clicking on the dropdown menu.

Created attachment 109035 [details]
Backtraces

I was able to reproduce this with two different backtraces.
I assigned the stabilizers to the CTRL + (1,2,3,4) shortcuts as per Isaac's comment, then switched between them while painting. It may take a bit, and you may need to release the mouse button and then start painting again a few times, but it does always seem to crash.

Sorry for the quick double post!

Selecting "Stabilizer" and then switching to any other smoothing type while drawing will almost always crash when the user stops painting, giving the first backtrace that I provided. This is the opposite order of what was described in original post in this report.
(The behavior of painting isn't correct in this state, either -- the cursor stops and lines are drawn from the real cursor position to the stuck cursor.)

It's alright man, I'm glad you double-posted. It shows that you discovered something.

I read your first comment then the second one, is there a workaround for this, or do I just have to stop the current brushstroke, change the stabilizer (1,2,3, or 4), then start again?

Also, I'd like to ask is this something that's still being looked into, or has it been confirmed that this is just simply how the program is?

(In reply to Isaac from comment #8)
> I read your first comment then the second one, is there a workaround for
> this, or do I just have to stop the current brushstroke, change the
> stabilizer (1,2,3, or 4), then start again?
Yes, I think you will have to wait, perhaps until the brushstroke has fully finished drawing on the screen, then start again.

> Also, I'd like to ask is this something that's still being looked into, or
> has it been confirmed that this is just simply how the program is?
I marked the bug as "confirmed," and the list of bugs marked that way could be considered a kind of to-do list.
I don't have the knowledge to fix it myself, but it may get fixed at any time by someone who does.
The good thing is that it doesn't crash only for you, so it should be a little easier for them to study what's wrong. :)

the easy way out is to check whether the stroke is still valid, instead of asserting on it:

void KisStrokesQueue::addJob(KisStrokeId id, KisStrokeJobData *data)
{
    QMutexLocker locker(&m_d->mutex);

    KisStrokeSP stroke = id.toStrongRef();
    Q_ASSERT(stroke);

The stroke is probably 0 because changing the smoothing option starts a new stroke.

Thread 1 "krita" received signal SIGSEGV, Segmentation fault.
KisStroke::lodBuddy (this=this@entry=0x0) at src/krita_3.3/libs/image/kis_stroke.cpp:301
301         return m_lodBuddy;
(gdb) bt
#0  KisStroke::lodBuddy (this=this@entry=0x0) at src/krita_3.3/libs/image/kis_stroke.cpp:301
#1  0x00007ffff38596ec in KisStrokesQueue::addJob (this=<optimized out>, id=..., data=0x86f2940) at src/krita_3.3/libs/image/kis_strokes_queue.cpp:344
#2  0x00007ffff38653a5 in KisUpdateScheduler::addJob (this=0x86bb108, id=..., data=data@entry=0x86f2940) at src/krita_3.3/libs/image/kis_update_scheduler.cpp:187

We already decided to change all asserts into KIS_SAFE_ASSERT too.

This should mitigate the issue:

diff --git a/libs/image/kis_strokes_queue.cpp b/libs/image/kis_strokes_queue.cpp
index a916b78472..23520f0104 100644
--- a/libs/image/kis_strokes_queue.cpp
+++ b/libs/image/kis_strokes_queue.cpp
@@ -345,7 +345,7 @@ void KisStrokesQueue::addJob(KisStrokeId id, KisStrokeJobData *data)
     QMutexLocker locker(&m_d->mutex);
 
     KisStrokeSP stroke = id.toStrongRef();
-    Q_ASSERT(stroke);
+    KIS_SAFE_ASSERT_RECOVER_RETURN(stroke);
 
     KisStrokeSP buddy = stroke->lodBuddy();
     if (buddy) {
@@ -364,7 +364,7 @@ void KisStrokesQueue::addMutatedJobs(KisStrokeId id, const QVector<KisStrokeJobD
     QMutexLocker locker(&m_d->mutex);
 
     KisStrokeSP stroke = id.toStrongRef();
-    Q_ASSERT(stroke);
+    KIS_SAFE_ASSERT_RECOVER_RETURN(stroke);
 
     stroke->addMutatedJobs(list);
 }
@@ -374,7 +374,7 @@ void KisStrokesQueue::endStroke(KisStrokeId id)
     QMutexLocker locker(&m_d->mutex);
 
     KisStrokeSP stroke = id.toStrongRef();
-    Q_ASSERT(stroke);
+    KIS_SAFE_ASSERT_RECOVER_RETURN(stroke);
     stroke->endStroke();
     m_d->openedStrokesCounter--;
 
lines 1-31/31 (END)


Note that I suspect that even with this patch, you will get really weird stuff happening, like a total restart of the stroke.

Yeah, still crashes, which is not too surprising.

(In reply to Boudewijn Rempt from comment #12)
> Yeah, still crashes, which is not too surprising.

Is there anything that I can do to help things along, or are things pretty much out of my hands at this point? (I'm open to anything that I can do to help.)

Well... The reason it crashes is that our stroke system isn't designed to suddenly start doing another type of stroke. So, basically, while painting, these shortcuts should be disabled. Going the other way, making this possible, is a huge refactoring. For now, I recommend not using those shortcuts while painting...

(In reply to Boudewijn Rempt from comment #14)
> Well... The reason it crashes is that our stroke system isn't designed to
> suddenly start doing another type of stroke. So, basically, while painting,
> these shortcuts should be disabled. Going the other way, making this
> possible, is a huge refactoring. For now, I recommend not using those
> shortcuts while painting...

I understand, and it seems if I stop my current stroke and then use a shortcut, there's no issue. (Compared to if I continue using the same stroke then activate a shortcut to change the stabilization.) I'm assuming that is fine, correct?

I didn't mean for this to turn into a long dig into code, rather I thought it was a bug. (Though, if others report it as well, I am curious what will become of this.) Thanks for helping me out with the issue, I really do appreciate it, especially all the developers/people who looked into this bug report. Much appreciated!

It definitely _is_ a bug. Krita shouldn't crash when pressing a key while in the middle of making a stroke.

I can confirm this happens also with with Krita 4.0.0-beta1.1.appimage (probably as expected due to the nature of the problem)

According to Boudewijn this has been accidentally fixed at some point and I am unable to reproduce it.

Really? That's awesome! Thanks for letting me know! :*)