Fixed bug #5799, Multires mesh modifications get reset by rendering
This bug can cause some nasty data loss (was introduces with my commit for using orco with multires.) Fixed by making sure to do multires_update_levels before applying modifiers.
have been in this particular case anyway), the blur radius will automatically
be limited to half the width or height of the image.
Nothing to do with this bug, but also now skips image region outside
borders when border rendering is enabled.
Crash entering faceselect mode in shaded mode.
The cause here is the new system that only preserves data through the
modifier stack as needed. The DerivedMesh and shaded colors could get
freed while still being used elsewhere.
Fixed bug #5756, Rendering artifacts when MRM is not set to maximum
Several changes were made:
* Added function multires_level_n to get the nth level from a multires mesh
* Removed the changes I made some time ago to init_render_mesh for multires meshes. Previously it was making a full copy of the mesh object in order to be able to apply deformations to the Pin level and propagate them to the Render level.
* Added two functions to DerivedMesh.c, multires_render_pin and multires_render_final. These two functions work together in the mesh_create_derived_*_render functions to apply all modifiers to the Pin level, then create the DerivedMesh from the Render level, and lastly restore the mesh to its original (undeformed) state.
* Added a check in multires_del_lower and multires_del_higher to ensure that level indices are properly clipped to the actual range of available levels.
Report had a very strange crash, which in the end appeared to be using
a filename of 165 chars long. This gave two bugs;
- the readBlog() function failed on that
- even worse: the struct Global still stored it in G.sce with 160 chars!
This is definitely an important issue... and (for the devs reading the
64 bits discussion) really not something allmighty coders are needed for!
:)
Composite: when using multiple scene render-nodes, and one of these
scenes got re-rendered (by making scene active temporary), the composite
cache should free the used buffers.
Now, on each render, all scenes in a Blender project are being checked.
Commit of patch #5385, to make forward cycling more
user-controllable. Previously it only worked on one axis,
which was auto-detected from movement. This allows forward
cycling to work in more situations, such as stair
stepping.
Order of freeing of Blender database is important, but the past year(s)
new database entries were added in wrong places, so freeing could go in
wrong order. In this case it was Groups being freed before Objects. But
there are more errors...
It only shows really in Windows, since it's the OS that is sensitive for
writing data in freed memory. For other OSes it doesn't make Blender more
stable, only mem-frees happened, no mem-allocs.
Compositing, depenency fix, so nodes execute only when really changed.
(Error was in nodes that don't get buffers as input, but only values. These
nodes flooded "execute" events through the node tree.)
Moved function used for 'Visual' keyframing, to live with the rest of the
API calls I added for Pose-Mode bone snapping. Now, it is in a more
'releasable' state.
- Bugfix #5714
New option "Save changed images" didn't respect relative file names
- Bugfix #5621
New camera (shiftx, shifty) was missing in IpoWindow as animatable option
Vertex color node worked only if VCol Paint/Light was enabled. Fixed
that, and removed the vertex color node making it part of the geometry
node instead.
Also, preview.blend had black vertex colors for the sphere, so set them
to white like the other primitives.
Rewrote the core-function responsible for ipo-cleaning. Now, it is in a less
wacko form. What is still not done is the conversion of a bunch of points
describing an arc to an arc defined by the handles of the keyframes on
either side of the arc. That will have to wait for the next development cycle.
This rewrite fixes two big bugs with the code:
* All but first curve got hidden aften cleaning
* Cleaning a curve with only two verts resulted in only one vert, even though
the values were not the same.
Changes to CustomData:
Some functions would only return the current active layer, added extra variants that take an index to select the level (modeled after CustomData_get_layer_n.)
Still todo:
* UVs are being interpolated linearly, should probably offer Catmull-Clark subdivision like Subsurf modifier.
* Vertex Colors still don't support multiple customdata layers.
* Editing UV data on levels other than 1 should be disabled in the interface (same for weights)
(Empty space will get OSA options, that I add tomorrow or so)
- Removed a lot of old unused variables in renderdata. Also meant I had
to remove this from python API... please check if this gives valid
scripts?
- Cleaned up bad formatted code for FFMPG buttons (spaces instead of tabs)
copy system with rigidbody constraint will link new constraints to new objects (instead of old)
fps from blender will be used when baking (left shift/ctrl/alt + P)
Crash leaving editmode related to vertex groups. I couldn't reproduce this,
but suspect somehow multiple deformvert layers were created. Added some
extra checks to avoid that now.
