Blender Git Loki

Fix T90651: camera reconstruction crash without scene camera

This was working differently in 2.79, tried tracking this down and it
seems this was wrong since the 2.8 beginning in {rB7907dfc40018}.

This would not only crash without an active scene camera, but would also
result in different tracks from different camera's constraints could not
be selected.

So select id depends on corresponding camera, remove the dependency on
scene camera completely.

Maniphest Tasks: T90651

Differential Revision: https://developer.blender.org/D12230

Cleanup: Spelling in Cycles X comments

Fix buffer size mismatch in SCRIPT_OT_python_file_run

Reading paths over 512 bytes would cause a buffer overrun.

Merge remote-tracking branch 'origin/master' into temp-lineart-contained

Cycles-X: Change viewport update system to be time based rather than sample based on fast devices

The viewport update speed when rendering with infinite samples has been changed to work based primarily on time since render start rather than sample count.

This change has been made because if you have a device fast enough to exceed 32 samples in 0.1 seconds, then the next viewport update you'll observe is 2 seconds after `render start` which can be annoying for anyone wishing to quickly iterate on fine details as `render start` is usually a 1 sample per pixel image meaning fine detail is lost in noise or overly blurred by a denoiser.

This change was made possible with help from Will Rusthon (a friend) and Blender developers.

Here is an example video of the issue (some of the "viewport changes" you see are due to compression artifacts to keep the video small):
{F10290321}

And here is my fix tested (the second demo with the caustics is there to show that the viewport updates reduce as time goes on. It's easier to see in the caustics scene due to the fireflies. This is important because reduced viewpoint updates means faster ray tracing. The difference is that with this patch the slow viewport update (once every 2 seconds) has been moved to after rendering for 8 seconds rather than occurring at 32 samples which could happen really quickly with simplistic scenes. Also the denoiser stopping at the end of the video has been fixed and is currently in the patch available for review):
{F10290508}

Reviewed By: sergey, leesonw

Differential Revision: https://developer.blender.org/D12265

Merge branch 'master' into subdivision_work

Sculpy dyntopo: fixed various topology bugs

* Fixed crash in dyntopo collapse. The
loops around vertex iterator dyntopo uses
doesn't actually work on non-manifold meshes,
or meshes with invalid normals, this was not
being checked in pbvh_bmesh_collapse_edge.
* Rotate tool now works with dyntopo.

VSE: Selection in preview area

Just basic select operator only with extend mode

This patch requires D12208, D12105 and D12320.

Differential Revision: https://developer.blender.org/D12343

Merge branch 'arcpatch-D12320' into transform-preview-select

Limit transform and gizmos to strips that are selected and visible at current frame

VSE: Use transform operator in preview

Initial implementation, not extensively tested, there may be bugs

ref T90156

Maniphest Tasks: T90156

Differential Revision: https://developer.blender.org/D12105

VSE: Add image transform gizmos

Patch based on D12105

Add transform tools and gizmos to sequencer preview area.

TODO:
- Caclculate center position for combined selection
- Limit selection in time
- Scale always works in local coords, but gizmo doesn't represent this
- Cage transform needs initial implementation for sequencer data

Differential Revision: https://developer.blender.org/D12208

Calculate gizmo center point and rotation.

Now I forgot that I did prior change here, so values were incorrect :(

Fix unsaved change

VSE: Use transform operator in preview

Initial implementation, not extensively tested, there may be bugs

Differential Revision: https://developer.blender.org/D12105

Change where new obj exporter writes temp files.

The code was writing relative to the test_release_dir flag,
but this didn't make a lot of sense.
Better to use Blender's tempdir mechanism, though since Blender
isn't fully initialized when using it for gtests, we have
to call the function to initialize the tempdir.

Refactor procedure building to add proper destructs, to use stacks

This still doesn't really work, but it solves a fundamental problem
with the order I was adding destruct calls for intermediate values.
The current problem is that the variables that a function depends
on (its inputs) are not added first, so basically a problem with the
traversal of the field network.

adaptive_cloth: AdaptiveMesh: setting EdgeData flags after split

Edges that had EDGE_BETWEEN_SEWING_EDGES when split, the new edges
should also be EDGE_BETWEEN_SEWING_EDGES.

adaptive_cloth: AdaptiveMesh: mark edges between sewing edges

Add a new flag for EdgeData that stores if the edge is between sewing
edges or not.

A function that marks all the edges that between sewing edges.

Call this function in the initialization of the static remeshing if
sewing is enabled.