Handle the possibility of a non-scalable font having the initial size by fluke and then failing on final size.

Much appreciated, thanks for the fix!

After some usage and feedback, the Footer's user experience is not good:

• It is difficult to discover.
• The Collapse/Expand arrow gets obscured by the scrollbar.
• There can be animation in other Spaces besides Animation Editors: the Compositor, Geometry Nodes, Shader Editor, etc..
• It's not much more convenient than dragging open a Timeline Space at the bottom of the Workspace (though it could be better with a hotkey toggle like the N-Panel/Sidebar).

I tested it again and now I can repro the problem. (Slot 1 cannot be removed)
I don't know how it worked before.

Still an issue.
Here a GIF showing the Deselect On Nothing option not working:

I don't think changing Window focus on hover is a good idea. The amount of unexpected issues this can cause will compound over time. It's also inconsistent with pretty much most other multi-window software out there. Issue with modifier keys not working on first interaction should be solved specifically, not with a monstrous solution that has many side effects.

@Germano Cavalcante (mano-wii) would you be able to check if this is still an issue? I'm not able to redo the double selection.

Slightly optimized cropping.

Marking one function argument as UNUSED.

Now not using FontBLF at all, only direct FreeType calls.

I think a complaint like this (and similar) would go away with this on Windows: https://developer.blender.org/D13951, since the target window would (most likely) be focused when moving the mouse there and would therefore get all keyboard input messages.

I can confirm the issue.
This seems to be a problem for any node editor, not just Geometry nodes.

Checked again in latest master build and I can't reproduce there.
Commit rB51381c94d8e0: BLF: Fallback Broken After Cache Removal has fixed the issue already. so I'm closing this report.
Don't hesitate to comment if problem remains in next daily build, I'll reopen your report.

Thanks for the report. I can confirm in latest master build. Raising the priority (will search for the breaking commit)

Harley: This looks like it can be done "bare metal", so don't use a FontBLF at all, open/close library, copy directly from FT_GlyphSlotRec->bitmap to supplied array. No locking needed. Should be as fast as could ever be and would not interfere if caching is added back in. Worth investigating anyway.

Ow, sorry. I was slightly modifying the task and didn't notice your changes. You are too fast :D

Can confirm (sort of).

In T100306#1401979, @Iliya Katueshenock (Moder) wrote:

I confirm. I suspect that this is due to the fact that the blender does not know what context to store information about the selected node tree behind? As if you deleted the selected object with material and it was not selected in the editor

Found some code that I think does a lookup on a node editor change. I'm not sure I'm well versed in the python api yet, but it looks like this doesn't take python into account node_context_path.cc

@Bastien Montagne (mont29) in this case couldn't the bounding box be increased for buttons where some extra wiggle room is acceptable? Then clipping can be kept strict without this being over zealous. I realize it's not exactly the same as the current behavior, but then at least it would show more characters without the ellipsis flickering in/out.

Hello @Hans Goudey (HooglyBoogly) , is this change convincing?

Not working correctly with the cache.

In the case of a font not returning an OS2 table, make sure it still uses the default latin string and not just blank.

Handle a font not returning a valid TT_OS2 font table.

@Yevgeny Makarov (jenkm) - It crashes on Mac, unfortunately.

• Updated to current master.
• Updated to use FreeType caching system
• Removal of user preference for latin sample
• New samples for Arabic and Devanagari
• blf_get_sample_text moved to blf_thumbs.c since not shared.
• IMB_thumb_load_font_get_hash moved back to current location in IMB thumbs_font.c

This seems like a good improvement, even if there are some minor trade-offs.

Thanks for the report. This looks expected from the present code.
Extend range (shift + LMB) property for outliner selection only extends selection from the active element to the cursor position.
TSE_ACTIVE flag is not set for the last selected element so active selection is not changed here.
Feature added in rBfdcae9ed0c78: Outliner: Add range extend select and modify keymap
Forwarding this report to developers because expected behavior is not mentioned (at least I could not find anywhere)

Could make some code optimization by just making K switch to the tool instead.
Would save lots of code left and right and group them at the same place. Worst case scenario make a new function to toggle back to the previous tool after the first manipulation.
Would prevent confusion too even I had this issue early on while learning blender, even today still I did not know it was 2 different tools, yet the same tool.

Hi Corbin, did you find a solution? Im facing the same problem with my Wacom in Windows Blender.

This still exists in 3.2.2.

@Krzysztof Trzciński (elmo) - Yes, added a return false after that BLI_assert_unreachable; I forgot that compilers might complain about that. LOL

With your patch and the extra return false it no longer crashes and I get a tooltip to appear both on the splash/greeting window and when hovering over icons in outliner.

All other places are followed by some kind of return, so I guess that answer that question. I think we want return false after the assert.

@Krzysztof Trzciński (elmo) - I wasn't able to add your name as reviewer to my diff (this interface doesn't seem to like your nick or name).

@Krzysztof Trzciński (elmo) - I can definitely try to compile with patch and see if it helps...

I can definitely try to compile with patch and see if it helps.
As for helping with understanding an issue and/or reviewing I am not sure how much I can help as I have very little familiarity with Blender source code and I have never before today looked at FreeType library.

@Krzysztof Trzciński (elmo) - It would be great if I could get your assistance with this. I might submit a patch and initially add just you as reviewer so that you can compile and confirm that it fixes this issue. Then I'd remove you and add a technical code reviewer. Does that sound like something you could do?

If landing a fix is problematic, reverting the caching commit that caused the issue should require much less approval?

The following might have wrong assumptions, so feel free to chine in, help out, and correct.

Looking at stacktrace I can see ui_but_tooltip_init and that hinted me on how to reliably reproduce issue. On splash screen I just move cursor over one of last open file names and wait. It seems that it crashes when it tries to show tooltip.
I have tried some other tooltip areas and oddly some work. I.e. thing in the properties panel seem to work. But hovering over "eye" icon or "camera" icon causes an ASAN crash.

I managed to get a crash with fast unwind disabled:

Same. this problem still persists.

This is a persistent issue that makes the shader edit very difficult to use.

• Use operator name for walk/fly mode

Update code-comments, use operator name for undo pushes.

Unfortunately I must confirm this. I'm not working on 'production' size scenes, yet with more complicated ones I find that I pause Cycles viewport, tweak some shader parameters and unpause the viewport to see the changes. It's not very efficient way of work,

Added more animators as reviewers of these changes.

In D15476#424346, @RedMser (RedMser) wrote:

I found the toggle for enabling/disabling the footer a bit hard to discover, but this seems to be a general UI issue in Blender so I assume this isn't something to be fixed for this patch specifically.