- User Since
- Nov 28 2009, 10:22 PM (437 w, 4 d)
Not sure why you need to do a list lookup. This is where LIB_TAG_COPY_ON_WRITE_EVAL comes in handy. But the tricky part is: this flags is *NOT* to be used for decision making, it is more a consistency check helper.
@John Luke Gibson (eaterjolly), can you compile in debug mode and run blender from gdb to get more meaningful backtrace?
Inverted return value by mistake
Some quick notes on progress.
Missed one commit in original patch
Building blender as module is not very official feature, and i'm not sure why logging will be attempted to be initialized twice. This is more a question to @Campbell Barton (campbellbarton).
@Campbell Barton (campbellbarton) , think it would make sense to wrap arguments of apply* functions into a context type of a structure (similar to ModifierUpdateDepsgraphContext). So then it's no longer an issue to modify arguments of those functions. Surely, some things would need to be still outside of structure (such as input mesh and output mesh/coords), but lots of other things can go there.
Not only this issue is already reported, system behaves according to its current design.
Tue, Apr 17
Avoid some extra NULL pointer checks
Went over regression tests and some Spring production files. Think there should be no stoppers from DEG to start porting operators.
Mon, Apr 16
From the description of the change -- +1.
Fri, Apr 13
Fix wrong deferred icon delete registration
Added strict flags to lockfree link list file
Update with fixes to atomic insert
As i understood, it's about BKE_scene_graph_update* functions, and NOT about modifier's eval/update functions. If it is so, then it's all correct approach.
Think is fine.
Thu, Apr 12
Please provide simple .blend file which demonstrates the issue.
Caused by rB651b8fb14eb.
Another report to consider when working on this feature: T54560.
Increasing SVM stack size will marginally increase GPU memory usage.
Tue, Apr 10
Can not reproduce the issue with copy-on-write enabled, which makes sense: visibility is flushed to the actual object, which causes conflicts between viewport and render, With copy-on-write enabled, both viewport and render has own data, without visibility flags conflicting.
Still can not reproduce, but i'm on Linux as well.
Thu, Apr 5
After all the discussion here at the studio, seems to be indeed less of an evil.