Issue seems to be that the node compiler handles custom sockets as floats. float sockets has a default_value attribute but is stored in a different way, hence an error occurs.

Use BM_iter_parallel.
Start counting from 0.

@Sebastian Parborg (zeddb) could help if you can find the commits. depending on how smooth it goes we might also consider a temp-branch.

Recent refactorings removed all the atomics but one in this case.

Number of atomics are reduced in master (only one)

• Use [0] to indicate that the struct has a dynamic size.
• Use min_iter_per_thread.
• Use different variable name for userdata.

Well they aren't really lengths or counts. in the end they are offsets. I found mat_offsets.tri_offsets a bit too long.

Resend diff after incorrect sending the previous time.

• Renamed SEQ_sequence_lookup_destruct to SEQ_sequence_lookup_free.
• Removed unused parameter in seq_sequence_lookup_new.
• Use BLI_ghash_str_new to create lookup hashmap, all names are zero-terminated.

• Revert "DrawManager: Cache material offsets."

• Renamed SEQ_sequence_lookup_destruct to SEQ_sequence_lookup_free.
• Removed unused parameter in seq_sequence_lookup_new.
• Use BLI_ghash_str_new to create lookup hashmap, all names are zero-terminated.

Due to recent changes is this still something we would want to integrate?
IMO the threading is good enough. Not sure if this fits in the parallel range implementation we have now.

I have no objections adding this. But would like to have feedback from fclem.

Just saw this fix, I committed a different one this morning. rB462bd8139937: Fix: Sequencer backdrop not updated during playback.

LGTM

Just tested and the known edge cases are working as expected. Only have minor comments that can be done before committing.

The reason for SEQ_sequence_removed(scene) is that multiple strips can be tagged for removal and would be removed at the same time. Updating the lookup per strip would need locking as setting a tag doesn't.
Per intent I do the updates only when accessing the lookup and never in between.

Removed commented out code.

• incorrect copyright in new files.

Although it sounds great to have a tagging system per item in the lookup, it is very hard to implement as you need to gather what has happening and in what order. When switching names of two strips before accessing them for example.

• Extracted lookup in its own file.

• Locking when rebuilding sequence lookup.

• Cleanup: rename to SEQ_sequence_find_by_name.
• Rebuild sequence lookup when strips are renamed.

• Added update_core_runtime functions to depsgraph.

• Flush caches when restoring the scene in depsgraph
• Added max key length to match sequencer->name.