My test file:

For the record, the API changed, see e7c3f7ba6f1385a2138862de8568e571fa97b5b4.

The original report is now fixed. Baking in 2.8 is still a bit strange/different when comparing to 2.7. That is a different issue than the reported one though.

For the records, this also causes T58917 to happen. I will see there if I can change the stored data to pointer (don't recall why I used strings in the first place).

Investigation update: bpy.context.scene.render.bake.cage_object is correct (e.g., "MyCage"), however when you clear it and reassign to the cage object you get leading spaces (e.g., " MyCage").

@Campbell Barton (campbellbarton) right, I suspected that much :) thanks for the updated fix.

Super confirmed, and the issue is a bit more strange in fact. I made an example that works in 2.79[*]:

@Campbell Barton (campbellbarton) for the records, should I have used message subscribing instead of a new listener?

@Sergey Sharybin (sergey) this file is much faster to crash reliably here (see weirdness 1 for instructions). It seems the same error as T58893.
Here at some point I got data llocated by T.17, freed by T.12, and with a read error by T.16,

I managed to reproduce a crash. Assuming is the originally reported:

Could you share a file in eevee where material index is used?
The compositing pass Material Index is not supported for EEVEE.

Cube is outside the collections. The object inside the collections is Cube.001 which is indeed hidden. I think you are mixing your objects?

I will look into this. It is strange because I'm almost sure I tested this on the original commit.

Committed on 80ffd7b4597.

That's all we need. @Clément Foucault (fclem) your rB0759480529c broke stuff.

This is more of a TODO than a bug. As you said yourself it has always been like this.
It is doable with a hack, using the WM_OT_call_menu operator to call the menu. Not sure if it makes sense though/

Ah, sorry I was a bit confused about the "worked" part of the report and scammed over.
Does it happen with --factory-startup as well?

Assigning to @Clément Foucault (fclem) although he is a bit skeptical of whether we can fix this since the shadow technique is not meant to work with those kind of (internally generated) triangles.
In fact I managed to reproduce this regardless of ngons:

Confirmed, and simplified the file to a minimum:

Can you go to Blender 2.7x, Help → Save System Info?
I'm particularly interested on the part about OpenGL:

OpenGL:
===========================================

In T58163#568321, @oleg krasnov (krasnovpro) wrote:

Indeed it seems to have been fixed.

Thanks for the extra files

Unless the issue only shows if the file was created with 2.8x, could you please attach a file that can be opened in 2.7x?
(and you can upload files here instead of hosting elsewhere)

And just for the records, up to 937136251aff9602be3d4b7b056edf43f6233328 could go into 2.8 with no problems. It is the equivalent of the "bring per-view layer collection visibility" with keeping eyes as they were.

Committed on a687d98e67823932f95bc3c1b1452109282a3e41.

I actually went out of my way to test it, and couldn't reproduce even on the revision you mentioned. Can you share the file right before you just pressed ta the first time?

Can you reproduce this from scratch in a latest blender or from a 2.7x file? Your file seems corrupted, as in, all metaballs show nothing even in object mode.
But I cannot reproduce it i a new file.

This is a known TODO, not really a bug, more some UX/UI polishing. @William Reynish (billreynish) wanna add this to the paper cut list?

I will test this next time we have the feature in blender. But at the moment collection visibility was temporarily reverted.

We need more information. Your graphic card for example, and whether blender 2.79 works in your computer.
If you launch blender from the CMD you may be able to read something after Blender closes.

Thanks for the report, but this was fixed already. Please test with the latest Blender and if you still have the problem I can re-open this.

Geez you are right, no idea why I passed something that was not a scene to the notifier. @Philipp Oeser (lichtwerk) your patch seems fine thanks for the fix. @Sergey Sharybin (sergey) only today I noticed you were the reporter ;)

For the records, I updated the sample files to svn:

@Philipp Oeser (lichtwerk) actually if the issue is probably either the message system (@Campbell Barton (campbellbarton)) or depsgraph (@Sergey Sharybin (sergey)).

Committed on 4c3ed98ca27667c3403361199096e31eaa93cce2.

Summoning @Campbell Barton (campbellbarton) here, in case he wants to take a look at it. Campbell I'm changing all the object iterators to take into account v3d.
(and just confirmed that indeed Cycles with renderer view still ignores local view)

The operator needs to be added back in the View menu still.

Done

• Add to View Menu
• Cleanup

Massive multi-object editing support

More fixes.

Addressed all the issues mentioned in review.
That said there are still the following known issues:

Latest blender2.8 HEAD