Age | Commit message (Collapse) | Author |
|
(See T48720).
|
|
This is in fact very hairy situation here... Objects are only refcounted by scenes,
any other usage is 'free', which means once all object instanciations are gone Blender
considers it can delete it.
There is a trap here though: indirect usages. Typically, we should never modify linked data
(because it is essencially useless, changes would be ignored and ost on next reload or
even undo/redo). This means indirect usages are not affected by default 'safe' remapping/unlinking.
For unlinking preceeding deletion however, this is not acceptable - we are likely to end with
a zero-user ID (aka deletable one) which is still actually used by other linked data.
Solution choosen here is double:
I) From 'user-space' (i.e. outliner, operators...), we check for cases where deleting datablocks
should not be allowed (indirect data or indirectly used data), and abort (with report) if needed.
II) From 'lower' level (BKE_library_remap and RNA), we also unlink from linked data,
which makes actual deletion possible and safe.
Note that with previous behavior (2.77 one), linked object would be deleted, including from linked data -
but then, once file is saved and reloaded, indirect usage would link back the deleted object,
without any instanciation in scene, which made it somehow virtual and unreachable...
With new behavior, this is no more possible, but on the other hand it means that in situations of dependency cycles
(two linked objects using each other), linked objects become impossible to delete (from user space).
Not sure what's best here, behavior with those corner cases of library linking is very poorly defined... :(
|
|
Also define single callback func typedef, cleaner this way!
Note: maybe we want to do that for the other callbacks too (data, etc.), but will be enough for now.
|
|
instanced operand
This is quite tricky situation which combined:
- Boolean modifier which accesses other object's derived mesh
(in fact, it's nothing to do with boolean modifier, any other
modifier which uses other's DM will have same bug).
- Dependency cycles in the scene which is rather russian-roulette
from the cycle solver point of view.
- Multiple instanced objects used as boolean operand.
With all this things combined boolean modifier was accessing operand's derived
mesh which was referencing data from Mesh datablock. The issue is that those
references are becoming invalid after EDBM_mesh_load().
This function already had code to make sure object itself does not end up with
dangling pointers from derived mesh. Make it now so no possible instanced objects
are left with dangling pointers.
And who said it's a good idea to reference something from derived mesh..
|
|
The Curves widget has buttons to zoom in on the curve. However the
click detection code doesn't take it into account, and at full zoom
in click on curve is detected very far from the actual visible curve.
Change it to compare the position to the actual line segments
in the UI coordinate space, i.e. with curve zoom applied.
|
|
|
|
This looks a bit odd, but being able to see if dyntopo is enabled
when the panel is collapsed is handy.
|
|
When no triangulation runs we can skip re-calculating normals.
|
|
This patch moves the PrintHardwareInfo() function in standalone only, not in embedded. Why? Because you can need this infos for debugging
purpose in "compiled" blender files but it was boring to have it displayed each time you launched embedded. So you can have this infos when you
click standalone or when you run your executable app from a console.
Reviewers: moguri, kupoman, panzergame.
|
|
|
|
Saves 8 bytes per vert/edge/face.
Gives overall ~20-25% memory saving for dyntopo sculpting
and modifiers that use BMesh.
|
|
|
|
|
|
changes in BLI_kdopbvh:
- `BLI_bvhtree_find_nearest_to_ray` now takes is_ray_normalized and scale argument.
- `BLI_bvhtree_find_nearest_to_ray_angle` has been added (use for perspective view).
changes in BLI_bvhutils:
- `bvhtree_from_editmesh_edges_ex` was added.
changes in math_geom:
- `dist_squared_ray_to_seg_v3` was added.
other changes:
- `do_ray_start_correction` is no longer necessary to snap to verts.
- the way in which the test of depth was done before is being simulated in callbacks.
|
|
Keep the fake header to ensure we always run.
|
|
|
|
|
|
|
|
|
|
0b5a0d84 caused regression since edit-mesh BVH wasn't cached,
each shrink-wrap modifier created its own BVH.
|
|
This way we avoid regression of sculpt mode shading.
Hopefully now it's all fine.
|
|
|
|
Regression caused by own changes to make texture paint more efficient
from workflow point of view.
Now the idea is to use vertex color outside of paint mode, so we don't
break any compatibility here.
|
|
|
|
For simple cases bitmasks were OK, but didnt work for vert/edge, vert/edge tests.
Tag verts instead, makes logic easier to follow and gives minor speedup.
|
|
Quiets assert
|
|
|
|
|
|
Reviewers: juicyfruit
Reviewed By: juicyfruit
Differential Revision: https://developer.blender.org/D1892
|
|
Make sure we don't perform any implicit address space conversion.
A bit annoying, but less intrusive approaches (like using temp private
variable in .cl kernel) do not work correct here.
Using generic address space will help from code side here, but will
be somewhat slower due to extra things happening as far as i know.
|
|
|
|
Removes many hash lookups per file-save and undo-step.
|
|
Other changes here planned which touch many lines, so run cleanup first.
|
|
|
|
Allows to plug/unplug different tablets while Blender is running.
Also fixes crash unplugging tablet while Blender runs (T48750).
|
|
|
|
In most the last added operator is being handles.
|
|
|
|
D2073 by @LazyDodo
|
|
|
|
So the error seems to be in cubic_tangent_factor_circle_v3(),
which was introduced with D2001.
I've tweaked the most obvious culprit here - the epsilon factor.
It used to be 10^-7, but I've reduced it down to 10^-5 now,
and it's looking a lot more stable now :)
---------
BTW, about the derivation of the magic 0.390464 factor I briefly subbed back
as a workaround for this bug, see:
http://www.whizkidtech.redprince.net/bezier/circle/
|
|
custom handle bones
It's probably some numeric precision issue, but until we figure out exactly what's
going wrong here, let's just revert back to the hardcoded value that was used here
successfully for years without issues.
|
|
Remapping indirect usage of IDs is forbidden from user space, this is calling for
nice nightmare with libraries handling (and undo crash, among other things).
Not sure why I was 'laxist' about indirect usage cases detection like that,
for now just consider any ID used by another linked datablock as indirect usage case!
Also, added some error/warning reports to Outliner's remap code.
|
|
The "W" channel will get a yellowish colour (i.e. a blend between the X/R and
Y/G axis colours), while the XYZ will behave as they do for other transforms.
|
|
The create drivers operator should not use the "Transform Channel" variable type
when driving one transform with another on the same object/bone. Otherwise, you
end up with a situation which technically results in a bit of a pseudo-dependency
cycle.
|
|
|
|
armatures
The Ctrl-Shift-C operator to add constraints between a pair of selected items,
for example, between two objects, or between two bones (in the same armature).
This commit makes it possible to use this operator to add a constraint where the
target is a bone from another object - e.g. to make a deform bone follow the control
bone in another armature, or to make an object use a bone as a tracking target.
Usage:
1) Ensure you are in Pose Mode, then select the bone to use as the target
2) Shift-Select the other object and/or the bone that's going to get the constraint
3) Ctrl-Shift-C
|
|
usage.
Also added warning to func doc, let's try to avoid this in future (for until we
pass string length systematically...).
|
|
(library) file.
Problem was in fact slightly wider, File space was nearly not taking into account
library navigation case and its 'virtual' directoris, except in a few places.
Add a wrapper around BLI_is_dir that also check for lib paths, and used it in
ED_file_change_dir(), such that we now always check path is a
valid directory - in the filebrowser context, not filesytem context. ;)
|
|
|