Fix crash when scrolling in smooth mode on empty thumbtable

[victoryforce]

Fixes #18357.

[TurboGit]

Thanks!

[TurboGit]

Merged in master & 5.0.x branch.

[victoryforce]

Regarding the release notes. An entry in the release notes should have been added if this crash was in version 5.0.0 and we fixed it. But it's unlikely that such an obvious problem could have been missed when testing the smooth scrolling PR. So I checked if 5.0.0 crashed and... no, it didn't. Running it under the debugger showed that in 5.0.0 when entering _event_scroll_compressed the thumbnail size (table->thumb_size) was non-zero even on an empty thumbtable.

So something changed between the release and a fairly recent development snapshot.

@TurboGit @jenshannoschwalm Does this change ring a bell to you? It seems that the zero as the thumbnail size on an empty thumbtable is a side effect of some recent change in the darktable codebase. Was this expected?

[jenshannoschwalm]

Unfortunately no.

[TurboGit]

No, no idea except maybe a change of Gtk+ version?

[victoryforce]

No, no idea except maybe a change of Gtk+ version?

I highly doubt that changing the library version could have such an effect. Everything is possible, but... My guess is that this is a side effect of some change in darktable code.

I'll bisect and find the culprit commit.

[victoryforce]

It seems that the zero as the thumbnail size on an empty thumbtable is a side effect of some recent change in the darktable codebase.

Yes, it was a side effect of the following commit:

c17601182375760792219d1a1b253fbad7ba8b9f is the first bad commit
commit c17601182375760792219d1a1b253fbad7ba8b9f
Author: Ralf Brown <[email protected]>
Date:   Sat Jan 18 14:09:25 2025 -0500

    rework splash screen to avoid keep-above

 src/common/darktable.c | 29 +++++++++++++++--------------
 src/control/control.c  |  8 ++++----
 src/dtgtk/expander.c   |  8 +++++---
 src/gui/gtk.c          | 56 +++-----------------------------------------------------
 src/gui/splash.c       | 17 +----------------
 src/gui/splash.h       |  2 --
 src/views/lighttable.c |  9 +++++----
 src/views/map.c        |  7 ++-----
 8 files changed, 35 insertions(+), 101 deletions(-)

It looks like we somehow lost a call to some initialization code that was responsible for calculating thumb_size with these changes. @ralfbrown Any ideas?

I don't think this is a serious regression, because although we ended up with a division-by-zero crash, it was actually a problem with the code that was dividing without proper safety checks. Rather, it is worth noting a certain architectural complexityy here if unrelated changes can have such unpredictable consequences.

[ralfbrown]

Lots of initializations are triggered by the initial realization of darktable's main window, which now happens much later.... Someone early in dt's development got overly clever with having GUI callbacks initiate things rather than simply making a direct call.