Blender Git Commits

Depsgraph: Get rid of simulation run left overs

Depsgraph: Get rid of depsgraph_util_rna.h

It only had two functions, half of which was never used, other half
was used only once. In any case, there's no actual reason to wrap
our RNA_access functions with just one more layer of abstraction.

Merge branch 'master' into depsgraph_refactor

Merge branch 'master' into depsgraph_refactor

Code Cleanup: Improved way that driver data gets printed for output

Don't include animation -> drivers link if there is no animation to require it

Depsgraph: Solution for missing update tag when adding new object

The issue was caused by new object add tagging object for recalc and tagging
relations for update. With new dependency graph it lead to situation when
ID tags are simply lost because of the whole graph being rebuilt.

The idea of this change is:

- Relations update tag doesn't destroy graph, it only stores flag in the
graph that it's relations need to be updated.

- ID tagging now stores an ID which is currently missing in the dependency
graph. Tag of those IDs would happen later after relations are all up to
date.

- Relations update reconstructs the graph and re-tags all the nodes which
were previously tagged for recalc.

This gives quite the same behavior that with old dependency graph, making
it possible to store recalc tags separately from the graph nodes themselves
in order to be able to invalidate graph without rebuilding it and still
have proper track for what's changed. This is a crucial thing to have in
order to have reasonable speed of python scripts.

Alternative to a separate storage in the Depsgraph would be to re-use
recalc flags stored in the ID itself.

There are probably some more tweaks to be done in this area, let's see
how this behaves for now.

P.S. This change also seems to solve issues with missing updates of the
armatures when exiting edit mode.

Depsgraph: Code and style cleanup

Mainly white spaces.

Depsgraph: Use explicit guarded allocation macros

This way we always are keeping track on memory usage of the depsgraph
(as opposite to previous: only when GXX guarded allocator is enabled,
which isn't enabled i.e. for release builds).

Allows us to track memory leaks earlier, plus allows us to keep an eye
on the memory usage of the graph itself.

Depsgraph: Inform when tagging didn't find node for tag

Also some code and style cleanup.

Depsgraph: Removing constraint should rebuild relations

TODO: We need to decide if we want to try supporting partial graph
updates or we'll just rebuild the whole graph when something changes.

Perhaps we'd better stick to the later for the simplicity and make
sure depsgraph build is not that slow.

Hunt for the missing relation updates continues.

Depsgraph: Changes to pose recalc tagging

Since the new dependency graph needs to have proper pose for building
dependencies the logic now is: if someone changes the pose he should
tag it for update, plus he should tag relations to be updated as well.

Depsgraph: Get rid of depsgraph_string_util

All the usages of the string_format could be replaces with
simple string concatinaiton. And further more, i'd prefer
using char* instead of strings to either names coming from
IDs or to statically allocated strings. Would be much better
from the performance point of view.

Depsgraph: Get rid of pose rebuild node

This way we request pose to be valid at the time of depsgraph build
and evaluation. Otherwise there's no clear way to pass individual
pose channels to the update callbacks.

Pose is going to be missing mainly on exit from edit more and file
load and seems this two cases are handled correct. At least assert
failure doesn't happen for me.

There would be some issues with linked koro which i've been fixing
in the master branch, this is to be investigated still.

Depsgraph: Add proper relations to the boolean modifier

This allows to get rid of some hardcoded relations between transform and geometry
components in the depsgraph itself.

Depsgraph: UNUSED() shouldn't be sued in headers

I'm not sure why it compiled last night and not today tbh.

Merge branch 'master' into depsgraph_refactor

add_relation() now prints errors everytime it can't find a required node

Judging by the results on Koro, we have a long way to go... :(

Fix for memory leaks and potential segfaults with invalid drivers

Depsgraph: debug_rebuild() will now actually rebuild the depsgraph