Fix errors thrown when picking terrain #3900
Conversation
…ce count to keep track of terrain vertex arrays.
👍 Totally agree with this (and the |
|
This fix seems awkward. Maybe it is the only approach, but can you walk me through it today along with the others that you tried. |
…ead or using reference counting.
|
@pjcozzi This is ready for another look. After our dscussion offline, I went back and changed it so that vertex arrays are queued to be destroyed at the start of the next frame. It only happens for tiles that are updated with new data. Vertex arrays of tiles that are being unloaded still happen immediately. It wasn't as bad as I thought. |
|
This is a bit cleaner and more direct. Update CHANGES.md. Is it possible to add a reasonable test? Or is it not worth it? |
|
@pjcozzi This is ready. I don't think I could reproduce it in a unit test. The only way I could reproduce the problem was to pick on mouse move and constant camera movement. |
As part of #3900, `vertexArraysToDestroy` was added as a parameter to `GlobeSurfaceTile.processStateMachine`, but none of the tests actually pass that parameter. This happens to be OK in master because none of the tests end up using the parameter, but in the 3D Tiles branch, the parameter is required and merging in master results in failing tests. Passing in an empty array fixes the issue (since we don't care about destroying the data in the specs), but we might possibly want to actually test that the parameter is being used correctly.
Fixes #3779.
The
GlobeSurfaceTileProviderkeeps a list of commands submitted for rendering. Let one of the commands contain a vertex array created from upsampling. After executing the draw calls but before rendering finishes, the tiles are processed (loading/creating/transforming imagery/terrain). During the processing, its possible that an upsampled terrain tile can receive its actual data. The upsampled vertex array will be destroyed and replaced with the new mesh. If a pick happens right after a render, the list of commands now has a destroyed vertex array from the upsampled mesh.There was a couple of fixes that I tried, but they all failed because some places in the code expect the commands executed the previous frame to be able to execute again. This was just the first place it would fail. The solution was to reference count the vertex arrays.
TLDR; We should reference count all WebGL resources.