4.3 KiB
Minutes from 42nd OpenVDB TSC meeting, Match 5th, 2020, (GMT)
Attendees: Nick A., Dan B., Jeff L., Ken M.
Additional Attendees: Bruce Chernia (Intel), Johannes Meng (Intel), JT Nelson (Blender)
Regrets: Peter C.
Agenda:
-
Confirm quorum
-
Secretary
-
Release notes build section
-
Update on Windows changes
-
Houdini -> VDB voxel size conversion
-
CI and Github Actions
-
LevelSet Tracking
-
Active value deep copy
-
Volume Advection grain size and topology options
-
Next meeting
-
Quorum was confirmed.
-
Secretary was Nick Avramoussis
-
Release notes build section
Proposal to separate out the build related release notes from the other release note sub-headings. It was noted that these notes have changed already in the past to adapt to new formatting being needed. Unanimous consensus, with the addition to additionally include any large changes in the highlights section. Nick to reformat the current release notes.
- Update on Windows changes
Nick has been looking through the Windows build and attempting to resolve various issues that have been raised. The following issues in particular are currently being addressed: #627, #624, #620, #611, #603 Main issues with the windows build continues to be the ability to support the both the Release and Debug builds in a single configuration as well as the both static and shared builds. vdb_view continues to be unsupported, but the work required to fix it is minimal. Ken to connect Nick to community who have reached out expressing desire to help improve the windows builds.
- Houdini -> VDB voxel size conversion
Nick has reported an issue to SideFX regarding conversion of Houdini DOP volumes to SOPs and subsequently to VDB volumes. The issue that manifests is that a volume which is intrinsically uniform in terms of its voxel size components can end up being represented as a non uniform scale map when converted to VDB. This can stop various optimizations from being performed further down the line and can additionally cause unnecessary re-sampling operations to occur. Jeff explained that Houdini volumes to not have an explicit representation of their voxel size. Instead, The voxel size of a Houdini volume is inferred from its transform projected onto a canonical unit cube. This can cause floating point differences to manifest when this value is calculated per component and presented to the user, as well during conversion to VDB. Nick mentioned that an ideal solution would be to be able to query some intrinsic details from the Houdini Volume's configuration and use this information to determine whether checks should be made on the final voxel size computations to ensure they are all equal. Jeff mentioned that Houdini Volumes currently contain no such information and instead proposed that a clamp can instead be used to ensure components match to a given tolerance. Note that whilst this problem originally focused on improving the conversion from Houdini->VDB, this issue could also be addressed by improving the comparisons between various VDB Map types. Nick to verify that the example provided to SideFX exhibits the above behaviour.
- CI and GitHub Actions
Dan has put a request in to the LF to disable appveyor and instead switch to github actions for our CI. Additional discussion around the new develop branch system. Jeff and Dan have both been caught out by having to switch the target branch to Master for TSC PRs. Nick suggested that we all should instead be merging into develop. Dan proposed an ad hoc target branch to instead be created for PRs which require additional changes from the TSC (coined "develop on demand"). Jeff, Ken and Dan voted in favour of this system vs maintaining a develop and master branch, with the condition a test is first demonstrated on an existing PR.
- LevelSet Tracking
- Active value deep copy
- Volume Advection grain size and topology options
Time
- Additional Discussion
Proposal to make IlmBase/OpenEXR optional currently in review but needs more changes. Questions were raised on why this would be useful, however various users of VDB seem to not require the writing and reading of half grids. Ultimately this further progresses the ability to build VDB with less enforced dependencies and makes it easier to build for users who do not need Half support.
- Next meeting
March 12th 2020. 2pm-3pm EDT (GMT-4).