Generic way for modifiers to check user-edited flags #108174

New Issue

Sergey Sharybin · 2023-05-23T12:02:05+02:00

Sergey Sharybin commented

2023-05-23 12:02:05 +02:00

The simulation modifiers need to take care of invalidating cache when user edits are performed on the modifier itself, or its dependencies.

For the point cache (rigid body simulation, for example) this is done as a dedicated dependency graph node, which is hooked to the rest of the graph via a relation which has RELATION_FLAG_FLUSH_USER_EDIT_ONLY flag. The benefit of such approach is that a generic callback function can be used (without capturing any internal state of the graph). The downside is the more complex routing of relations.

The simulation nodes took a bit different approach: the nodes modifier has a special lambda callback which checks the node and dependency graph state:

modifier_node->evaluate = [](...) {
  if (modifier_node->flag & DEPSOP_FLAG_USER_MODIFIED) {
    ...
  }
}

Such approach allows to access the modification state without extra nodes and relations, but it is not very generic: ideally the modifier-specific logic will be left to the modifier's "apply" callback.

Fortunately, there is a simple modification to the current nodes modifier callback we can do!

General idea: have a generic modifier node callback (for all node types) which will copy the user-modified flag from the modifier dependency graph node to the modifier data. The modifier's "apply" callback then can make decision based on this flag.

Pseudo-code of this proposal goes as following:

modifier_node->evaluate = [modifier_index = ...](::Depsgraph * /*depsgraph*/, ...) {
  ModifierData *md_eval = object_eval->modifiers[modifier_index];
  if (modifier_node->flag & DEPSOP_FLAG_USER_MODIFIED) {
    md_eval-> flag |= MODIFIER_USER_MODIFIED;
  } else {
    md_eval-> flag &= ~MODIFIER_USER_MODIFIED;
  }
}

Use the modifier index since it can be captured at the graph construction time, and will stay available even if the original objects are destroyed (the long-term goal is to allow the depsgraph be fully independent from the original scene after its Copy-On-Write was run). It is a bit annoying to do an index-based lookup in the list, but it is unlikely to show up in a profiler.

The geometry node modifier can immediately use this new flag. In a longer term the rigid body will do the same or be migrated to the new simulation framework :)

The simulation modifiers need to take care of invalidating cache when user edits are performed on the modifier itself, or its dependencies. For the point cache (rigid body simulation, for example) this is done as a dedicated dependency graph node, which is hooked to the rest of the graph via a relation which has `RELATION_FLAG_FLUSH_USER_EDIT_ONLY` flag. The benefit of such approach is that a generic callback function can be used (without capturing any internal state of the graph). The downside is the more complex routing of relations. The simulation nodes took a bit different approach: the nodes modifier has a special lambda callback which checks the node and dependency graph state: ```Cpp modifier_node->evaluate = [](...) { if (modifier_node->flag & DEPSOP_FLAG_USER_MODIFIED) { ... } } ``` Such approach allows to access the modification state without extra nodes and relations, but it is not very generic: ideally the modifier-specific logic will be left to the modifier's "apply" callback. Fortunately, there is a simple modification to the current nodes modifier callback we can do! General idea: have a generic modifier node callback (for all node types) which will copy the user-modified flag from the modifier dependency graph node to the modifier data. The modifier's "apply" callback then can make decision based on this flag. Pseudo-code of this proposal goes as following: ```Cpp modifier_node->evaluate = [modifier_index = ...](::Depsgraph * /*depsgraph*/, ...) { ModifierData *md_eval = object_eval->modifiers[modifier_index]; if (modifier_node->flag & DEPSOP_FLAG_USER_MODIFIED) { md_eval-> flag |= MODIFIER_USER_MODIFIED; } else { md_eval-> flag &= ~MODIFIER_USER_MODIFIED; } } ``` Use the modifier index since it can be captured at the graph construction time, and will stay available even if the original objects are destroyed (the long-term goal is to allow the depsgraph be fully independent from the original scene after its Copy-On-Write was run). It is a bit annoying to do an index-based lookup in the list, but it is unlikely to show up in a profiler. The geometry node modifier can immediately use this new flag. In a longer term the rigid body will do the same or be migrated to the new simulation framework :)

Sergey Sharybin added the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Generic way for modifiers to check user-edited flags #108174