This is a workaround for T80804.
There's a startup crash that happens on 2.91.0 on Windows, an `EXCEPTION_ACCESS_VIOLATION`
on `atio6axx.dll`. It is triggered by `glClear` on the `detect_mip_render_workaround`
function. The workaround moves the function after the device/driver workaround section and
sets the flag to the affected one to avoid running the check.
It is deprecated hardware that has not meet the minimum requirements since 2.79, but is
still usable and this extends its usability a bit before the cards are finally blacklisted.
Reviewed By: Jeroen Bakker
Differential Revision: https://developer.blender.org/D9667
This reverts commit cbae82ba96.
This change introduced the following problems:
- We could no longer reliably duplicate or use an existing
custom property names.
- We could no longer assume a bone or ID name
can be used in a custom-property.
- Importers that support custom properties (such as FBX)
could fail with an exception creating custom properties.
This was added when Linux's audio support
often needed to be manually configured.
Further 3rd party libraries have their own environment variables & docs
these need not be part of Blender's help message unless they're likely
to be needed to properly run Blender.
Implement improvement from T73139 for merging along edges.
It is now called "Connected" mode, while the default is called "All".
With the recent performance improvement, the Connected Mode is in some
cases only double the speed than the usual merge all strategy but in
other cases it may be even faster. The bottleneck is somewhere further
down the line of merging geometry.
The motivation for this patch came from T80897, because the merging in
complex solidify is making it very slow.
Now merging can be removed from solidify without greater consequences,
as this is just a quicker and more advanced algorithm to do the same
thing that solidify currently does slowly.
Reviewed by: mano-wii, campbellbarton
Ref D8966
When opening a temporary File Browser, we have to make sure the file selection
parameters are refreshed. When opening it in a new Window that would always be
the case, if the File Browser uses a maximized window (as set in the
Preferences), it might reuse space-data from a previous use. So we have to
force the refresh.
Also renamed the relevant function to be more clear about what it's doing.
Mistake in 95b3c4c966.
This just makes sure that the mesh coming out of the subdivision
surface node has correct normals. Ideally, we would lazily compute
derived data like normals in more cases, but that can be solved later.
Correctness is more important right now.
In order to solve this better, it would be nice if functions like
`BKE_mesh_ensure_normals` would take a `const Mesh *`.
The mesh could be considered to be logically const, because
normals are derived data that is only cached for the current
state of the mesh.
Buffer strokes weren't being the excluded from depth only draw calls
so were being included in depth tests. They are now excluded by
bypassing the creation of the buffer strokes.
Reviewed By: fclem
Differential Revision: https://developer.blender.org/D9742
* Avoid direct access to `SpaceFile.params`, use a getter instead. This matters
because once the asset-browser changes are in, there will be an alternative
selection parameter object. The getter can return the correct one.
* Rename the function to ensure the parameters. The old name
`ED_fileselect_get_params()` wasn't a mere getter, it would create the
parameters if necessary. Now we have an actual getter, so better be clear.
* In some instances, I replaced the old "get" function with the new mere
getter. So the ensure logic is called less often. However, in these cases we
should be able to assume the selection parameters were created already as
part of the editor creation routine.
The term "active" in the new function names may seem a bit odd in the current
context, but that is a preparation for the Asset Browser merge as well. Like
said, there will be two file selection parameter objects in the space.
Caused problems in the Asset Browser branch when passing an empty path. The
function shouldn't blindly add a slash but sanity-check the input parameters
first.
Also makes NLA tracks and strips overridable.
User can either edit existing strips in existing NLA tracks (but not add or remove them), and/or add new NLA tracks after those comming from the linked data.
Most of the work was as usual checking operators and adding protections against illegal operations in override context.
Note that since we can only rely on indices to deal with local added tracks, we forbid any local track being before any linked/original track.
Maniphest Tasks: T72629
Differential Revision: https://developer.blender.org/D9611
It's easier to read and less 'weird' to check that an item is non-local
in a liboverride data-block, than the other way around. Thanks to
@sybren for noticing it.
The transform modes `shrinkfatten` and `seq_slide` have a special way of
handling events.
They use modal events in a different way than expected.
Therefore, this commit adds special event handles for these modes and
removes the keymodal tips from the status bar.
These effects are already described in the header anyway.
If the stroke had zero points the pointer returned NULL.
Also replaced for loop by FOREACH macro.
This is a corner case of "empty" strokes without points.
Change the default for the Show Channel Group Colors preference to OFF.
This option in the user preferences was introduced in rBad85256e7108. It
moved a per-file-per-editor option to the user preferences. As this
option would be frequently turned off by animators, this would now have
to happen only once. This commit takes this one step further, and turns
it off by default, as it can cause major readability issues of the
animation channel list.
RNA collections that support insertion of new items in liboverride
data-block need a special way to distiguish between locale and
orig-from-linked items (since some operations are allowed on the forer,
but no the latter).
In future we want a proper solution to abstract that at the
`BKE_lib_override` level, but for now we need to add some code for each
case.
Note that this commit also fixes a few potential issues with GPencil
modifiers, and constraints, regarding their handling of local overrides.
Fix the Assign Automatic Weights operator in weight paint mode when the
Vertex Groups X option is enabled. This issue was probably introduced in
rB5502517c3c12 where `me->editflag & ME_EDIT_MIRROR_X` was replaced by
either `me->editflag & ME_EDIT_VERTEX_GROUPS_X_SYMMETRY` or
`me->symmetry & ME_SYMMETRY_X`. In this case, the former wasn't working,
so I replaced it with the latter.
Continuation of the work started with 249e4df110. After all display modes
were ported to this new design, this commit starts the (more complex) work on
the individual tree-element types. More concretely it ports animation
tree-elements (action data-blocks, drivers and NLA data).
The commit above explains motivations. In short, we need a better design that's
easier to reason about and better testable.
Changes done here are pretty straight forward and introduce similar class
hierarchy and building patterns as introduced for the display modes already.
I.e. an abstract base class, `AbstractTreeElement` with derived classes for the
concrete types, and a C-API with a switch to create the needed objects from a
type enum. The latter should be replacable with something nicer later on (RAII
based, and type-safer through meta-programming).
Each tree-element type has its own class, with an own header and source file
(okay some closely related types can share a header and source file, like the
NLA ones).
I added some further temporary bits for the transition to the new design, such
as the `TreeElement.type`. It should entirely replace `TreeElement` eventually,
just as `outliner_add_element()` should be quite small by then and easily
replacable by a `TreeBuilder` helper.
Mistake in 35a5dee2ef.
Code would simply assume that the element under the cursor is an Object if it
was an ID (but not a collection).
This wouldn't cause any issues in current code, since the only other ID that
set the direct-data were collections, which are special in that they don't have
a `TreeStoreElem.type` of 0, which is already being checked for here. And if
there was no direct-data set, the object lookup in the View-Layer would
correctly fail too.
I carefully checked and am quite sure for `TSE_ANIMATION_DATA` and
`TSE_NLA_ACTION` the direct-data isn't used at all. This data is stored and
accessed in unsafe ways based on conventions anyway (`void *`). We should aim
for a safer design but it will be difficult to get there. Any removed
complexity will help.
Non-ID tree-elements would cast their data pointer to `ID *` and take the name
and ID-Code from there. The name would actually be overridden a few lines
later, so that didn't cause issues. But the ID-Code stored inside the tree
element kept an undefined value. In practice that probably didn't cause many
issues either, since it's just an undefined value that was very unlikely to
take a valid 16-bit ID-code value, meaning ID-Code checks would simply fail as
they should. Further there typically are other checks to see if the element
actually represents an ID.
However, in theory this may have caused a few random crashes or
data-corruptions.
Modernize loops by using the `for(type variable : container)` syntax.
Some loops were trivial to fix, whereas others required more attention
to avoid semantic changes. I couldn't address all old-style loops, so
this commit doesn't enable the `modernize-loop-convert` rule.
Although Clang-Tidy's auto-fixer prefers to use `auto` for the loop
variable declaration, I made as many declarations as possible explicit.
To me this increases local readability, as you don't need to fully
understand the container in order to understand the loop variable type.
No functional changes.
