5.0 KiB
Minutes from 16th OpenVDB TSC meeting, May 9, 2019
Attendees: Ken M., Nick A., Dan B., Peter C., Jeff L.
Additional Attendees: Bruce Chernia (Intel), John Mertic (LF)
Agenda:
-
Confirmation of quorum
-
Selection of secretary
-
6.1 release
-
Plans for next release
-
Siggraph Course
-
CI Update
-
SOP OpHiding - OVDB-64 / PR404
-
Schedule next meeting
-
A quorum was confirmed
-
Secretary - Dan Bailey
-
6.1 Release
Peter has completed the 6.1.0 release, Ken to announce the release on the Google Forum.
- Plans for next release
Next minor release scheduled for around Siggraph, ideally before but preferably no later than August to allow SideFX time to integrate it into the next release of Houdini. Intention for this next release to include a number of big changes - AX from Nick, tools previously shared via Google Doc from Ken, delayed-loading and improved VDB points advection from Dan, generic convolution and sharpening tools from Peter. We are hoping for progress towards receiving a contribution from Autodesk too. The following release will be a major ABI one in Q3/Q4.
There was discussion around using feature branches that live in the main repository to allow large new contributions such as AX and Multi-Res to mature. Proposal is to use PRs in the normal way and to continue using one approval on every PR for the time being. No opposition to this.
- Siggraph Course
There appears to have been a change in how Course Notes are shared at Siggraph requiring them to have been prepared by 14th May this year, which is significantly sooner than previous years. Jeff suggested Ken reach out to Michael Reed as the Courses Chair for clarification. Nick and Dan to share slide decks with Ken from previous years to help with preparing material.
- CI Update
With integration of CMake, migration to Circle CI from Travis CI has been completed. Two outstanding issues, a segmentation fault that appears to happen sporadically, most commonly on the ABI=4 branch. Dan has recently merged a PR to migrate the ABI=4 branch from TBB 4.4.0 to 4.4.6 which seems to have improved stability. Need to wait a little longer to fully assess the impact. A build failed in the ABI=5 build today so question about whether this may have been related to TBB. Nick suggested perhaps it was related to memory allocators. Ken suggested we enable asserts in release mode. The second issue is that the Circle CI configuration is not setup correctly which manifests in an issue where the builds are kicked off from a PR but not linked back to it rendering the PR un-mergeable. Dan has emailed Andrew Grimberg at the Linux Foundation about resolving the issue and is waiting to hear back. Circle checks have been downgraded to not required for the time being.
- SOP OpHiding - OVDB-64 / PR404
Artists are regularly confused about having duplicate VDB/OpenVDB SOPs show up in the tab menu in Houdini when installing the open-source SOPs. Peter says at Dreamworks, many shows use opcustomize scripts to alleviate this issue, Nick reports that DNeg is also using opcustomize scripts.
The committee was largely split on how best to solve this problem. Dan and Jeff have discussed one proposal and Dan has submitted as PR OVDB-64 which Jeff has approved. Peter has concerns with this approach in that there would be two contradictory mechanisms for hiding nodes, one using the SideFX ophide script mechanism and one using a compiled ophide policy mechanism and would prefer to see them unified. Jeff highlighted an example scenario where a facility may wish to hide all of the open-source SOPs except for one which they had modified and extended. The solution to this scenario under this proposal would require modification of the SOP definition itself by either manipulating the existing policy flags or adding a new flag which would need support added to all existing SOPs. Otherwise, the facility would fall back on using an opcustomize script.
The simple alternative to this proposal is a single script that has all the ophide commands in it to allow users to customize as they chose and then install it. However, there are a number of key problems with this approach. It requires users to deploy a new configuration file which adds an extra source of unreliability in deployment. It doesn't support users installing only a subset of the nodes without running the risk of neither SOP being visible. It requires contributors to the project to maintain a side-car file in addition to the existing SOPs. Nick and Peter briefly discussed auto-generating an opcustomize file using CMake but not yet clear how that might work.
The meeting wrapped up without a clear consensus on the way forward on this issue. Dan mentioned that Siggraph would be a nice opportunity to share a solution with the community if possible and invited others to contribute an alternative implementation for discussion in a future meeting. The Committee is also looking to gather more feedback from production to better inform this decision.
- Next Meeting
May 23rd 2019. 3pm-4pm EDT (GMT-4).