The issue was discovered only after recent changes, but roots back
to much older changes.
What was happening is scene's ID recalc flags where never cleared,
which caused ensure_view_layer() to always run copy-on-write on the
scene. This resulted in certain runtime data being cleared, without
proper flag stored in the dependency graph.
This was caused by ID recalc clear function checking whether any ID
was tagged for recalc in that graph or not. This was happening due
to all areas using DEG_id_type_tag() which can only set flags on the
graph from viewport scenes, and could not inform render dependency
graph.
Now ID tyoe tagging is happening on per-graph level, which avoids
possibility of flags running out of sync.
In a bit longer term we also need to get rid of two functions which
are clearing flags: DEG_id_type_tag() and deg_graph_clear_tags().
BF-admins agree to remove header information that isn't useful,
to reduce noise.
- BEGIN/END license blocks
Developers should add non license comments as separate comment blocks.
No need for separator text.
- Contributors
This is often invalid, outdated or misleading
especially when splitting files.
It's more useful to git-blame to find out who has developed the code.
See P901 for script to perform these edits.
Some summary of changes:
- Don't use DEG prefix for types and enumerator values:
the code is already inside DEG namespace.
- Put code where it locally belongs to: avoid having one
single header file with all sort of definitions in it.
- Take advantage of modern C++11 enabled by default.
There were at least three copies of those:
- OB_RECALC* family of flags, which are rudiment of an old
dependency graph system.
- PSYS_RECALC* which were used by old dependency graph system
as a separate set since the graph itself did not handle
particle systems.
- DEG_TAG_* which was used to tag IDs.
Now there is a single set, which defines what can be tagged
and queried for an update. It also has some aggregate flags
to make queries simpler.
Lets once and for all solve the madness of those flags, stick
to a single set, which will not overlap with anything or require
any extra conversion.
Technically, shouldn't be measurable user difference, but some
of the agregate flags for few dependency graph components did
change.
Fixes T58632: Particle don't update rotation settings
With the current implementation it only confuses logic around
checks like vewport/render subdivision levels.
If this mode is really needed for any decision making, implement
this properly.
Among all the lines moved around, the general idea is quite simple.
Actually, there are two ideas implemented there.
First one, is when object itself is tagged for update, we tag its
point cache component for evaluation, which makes it so point cache
is properly reset. We do it implicitly because otherwise we'll need
to go everywhere and add explicit tag in almost all the properties.
Second thing is, we link all collider and force fields to a point
cache component using special type of link. This type of link only
allows flush if change is caused by a user update. This way reset
does not happen when change is caused due to animation, but will
properly happen when user causes indirect change to the objects
which are part of physics simulation.
The Nearest Surface Point shrink method, while fast, is neither
smooth nor continuous: as the source point moves, the projected
point can both stop and jump. This causes distortions in the
deformation of the shrinkwrap modifier, and the motion of an
animated object with a shrinkwrap constraint.
This patch implements a new mode, which, instead of using the simple
nearest point search, iteratively solves an equation for each triangle
to find a point which has its interpolated normal point to or from the
original vertex. Non-manifold boundary edges are treated as infinitely
thin cylinders that cast normals in all perpendicular directions.
Since this is useful for the constraint, and having multiple
objects with constraints targeting the same guide mesh is a quite
reasonable use case, rather than calculating the mesh boundary edge
data over and over again, it is precomputed and cached in the mesh.
Reviewers: mont29
Differential Revision: https://developer.blender.org/D3836
When this works correctly, we should be able to feed in an existing
depsgraph instance, and get out a "filtered" copy of it that contains
only the subset of nodes needed to evaluate what we're interested in.
The current implementation only filters on ID blocks/nodes,
and starts by building a full new depsgraph instance first.
I'd originally intended to do it per operation instead, copying
over individual nodes as appropriate to have the smallest and least
memory intensive graph possible. However, I ended up running into
into problems with function binding + COW arguments, hence the
current slow solution.
When active dependency graph is evaluated, it will apply animation,
drivers and scalar evaluation data (such as object matrix) to an
original datablock. This way operators and tools can easily read
data from original datablock.
This will simplify porting them to copy-on-write, and solve issues
when some operator will allocate new datablock based on original one,
and will want to read data from it.
After discussion with Sergey and Dalai, we have decided to remove
this option completely. We're getting to the point where it is almost
impossible to really use 2.8 without COW, and keeping the old option
running ends up diverting dev resources away towards tracking down
and fixing problems with a parallel system that will be going away.
This introduces a new depsgraph API for getting updated datablocks,
rather than getting it from bpy.data.
* depsgraph.ids_updated gives a list of all datablocks in the depsgraph
which have been updated.
* depsgraph.id_type_updated('TYPE') is true if any datablock of the given
type has been added, removed or modified.
More API updates are coming to properly handle multiple depsgraphs and
finer update granularity, but this should make Cycles work again.
As was decided at today's dev kickoff, we're now moving to having
Copy-on-Write enabled by default, as 2.8 is barely functional with
it off.
To run Blender *without* COW (e.g. for testing), use:
--disable-copy-on-write
OVERVIEW
* In 2.7 terminology, all layers and groups are now collection datablocks.
* These collections are nestable, linkable, instanceable, overrideable, ..
which opens up new ways to set up scenes and link + override data.
* Viewport/render visibility and selectability are now a part of the collection
and shared across all view layers and linkable.
* View layers define which subset of the scene collection hierarchy is excluded
for each. For many workflows one view layer can be used, these are more of an
advanced feature now.
OUTLINER
* The outliner now has a "View Layer" display mode instead of "Collections",
which can display the collections and/or objects in the view layer.
* In this display mode, collections can be excluded with the right click menu.
These will then be greyed out and their objects will be excluded.
* To view collections not linked to any scene, the "Blender File" display mode
can be used, with the new filtering option to just see Colleciton datablocks.
* The outliner right click menus for collections and objects were reorganized.
* Drag and drop still needs to be improved. Like before, dragging the icon or
text gives different results, we'll unify this later.
LINKING AND OVERRIDES
* Collections can now be linked into the scene without creating an instance,
with the link/append operator or from the collections view in the outliner.
* Collections can get static overrides with the right click menu in the outliner,
but this is rather unreliable and not clearly communicated at the moment.
* We still need to improve the make override operator to turn collection instances
into collections with overrides directly in the scene.
PERFORMANCE
* We tried to make performance not worse than before and improve it in some
cases. The main thing that's still a bit slower is multiple scenes, we have to
change the layer syncing to only updated affected scenes.
* Collections keep a list of their parent collections for faster incremental
updates in syncing and caching.
* View layer bases are now in a object -> base hash to avoid quadratic time
lookups internally and in API functions like visible_get().
VERSIONING
* Compatibility with 2.7 files should be improved due to the new visibility
controls. Of course users may not want to set up their scenes differently
now to avoid having separate layers and groups.
* Compatibility with 2.8 is mostly there, and was tested on Eevee demo and Hero
files. There's a few things which are know to be not quite compatible, like
nested layer collections inside groups.
* The versioning code for 2.8 files is quite complicated, and isolated behind
#ifdef so it can be removed at the end of the release cycle.
KNOWN ISSUES
* The G-key group operators in the 3D viewport were left mostly as is, they
need to be modified still to fit better.
* Same for the groups panel in the object properties. This needs to be updated
still, or perhaps replaced by something better.
* Collections must all have a unique name. Less restrictive namespacing is to
be done later, we'll have to see how important this is as all objects within
the collections must also have a unique name anyway.
* Full scene copy and delete scene are exactly doing the right thing yet.
Differential Revision: https://developer.blender.org/D3383https://code.blender.org/2018/05/collections-and-groups/
The depsgraph was always created within a fixed evaluation context. Passing
both risks the depsgraph and evaluation context not matching, and it
complicates the Python API where we'd have to expose both which is not so
easy to understand.
This also removes the global evaluation context in main, which assumed there
to be a single active scene and view layer.
Differential Revision: https://developer.blender.org/D3152
This was only used for viewport rendering, where we can just pass the engine
type directly. There is no technical reason why we can't draw the same depsgrpah
with different render engines.
It also led to some weird things like requiring a render engine for snapping
and raycast API functions.
Differential Revision: https://developer.blender.org/D3145
Scene, view layer and mode are now set in the constructor and never changed.
Time is updated on frame changes to indicate which frame is being or has been
evaluated last.
This is a step towards making EvaluationContext obsolete.
Differential Revision: https://developer.blender.org/D3144
Now the only missing bit seems to be in Cycles to pass depsgraph to
builtin_image_float_pixels().
Ideally we could get evaluation context instead of using depsgraph + settings.
But for the other rna EvaluationContext functions this is how we are doing.
Reviewers: sergey, brecht
Differential Revision: https://developer.blender.org/D3087
User notes
----------
Compositing, rendering of multi-layers in Eevee should be fully working now.
Development notes
-----------------
Up until now we were still using the same depsgraph for rendering and viewport
evaluation. And we had to go out of our ways to be sure the depsgraphs were
updated.
Now we iterate over the (to be rendered) view layers and create a depsgraph to
each one, fully evaluated and call the render engines (Cycles, Eevee, ...) with
this viewlayer/depsgraph/evaluation context.
At this time we are not handling data persistency, Depsgraph is created from
scratch prior to rendering each frame. So I got rid of most of the partial
update calls we had during the render pipeline.
Cycles: Brecht Van Lommel did a patch to tackle some of the required Cycles
changes but this commit mark these changes as TODOs. Basically Cycles needs to
render one layer at a time.
Reviewers: sergey, brecht
Differential Revision: https://developer.blender.org/D3073
Those are unused, and not clear whether we will ever support this.
Seems to be better having more like "component" tags, would be less magic
involved to guess what exactly is to be tagged.
This commit effectively reverts fix T45702 done in 067fe2719a.
Reasoning:
- Blender Internal is being replaced with Eevee, and will be removed entirely
rather soon.
- All render engines are planned to have own depsgraph, so such threading
conflicts should no longer be an issue.
- We don't want to spend time on porting workarounds for EOL things to a new
design. Less code -- faster the work :)
- If such notifications will end up needed for some other cases, we would
need to re-implement this a more proper depsgraph tagging/flushing and make
it to work with all copy-on-write datablocks and everything.
