Can you provide a simple test .blend file that shows the bug?
Right now it's only instantiated for different kinds of inputs. But it's not impossible to also instantiate for different kinds of outputs if that ever becomes necessary.
I think it's perfectly fine to do that right now to avoid all the boilerplate. We could potentially investigate optimizations that avoid the computation of unnecessary outputs if necessary.
Looks pretty good and works well in my test with the supported nodes.
Make sure the include is in all updated node files, unity builds might hide missing includes.
It has not been formally agreed upon yet. I'm just pushing for it a bit. Also a few things have changed since the original implementation. E.g. we can properly support panels in the modifier now.
ASAN Report
==182792==ERROR: AddressSanitizer: heap-use-after-free on address 0x7f11d9ac5800 at pc 0x000003062d22 bp 0x7ffcff898870 sp 0x7ffcff898868
READ of…I can reproduce it. It crashes earlier already in a debug build.
Backtrace
__pthread_kill_implementation@0x00007fffe64ae834 (Unknown Source:0)
raise@0x0000…If not filtering by type, maybe we can modify
DEG_OBJECT_ITER_BEGINinto using object_dependencies list as a reference and only return objects that has the original id of those in that list?…
Haven't looked at the code in too much detail yet. It seems the behavior is still different compared to legacy curves.