TSC Meeting Notes 2021-08-26

Owned by Cary Phillips
Last updated: Aug 26, 2021
2 min read

Attendance:

Cary Phillips
Christina Tempelaar-Leitz
John Mertic
Joseph Goldstone
Kimball Thurston
Larry Gritz
Nick Porcino
Peter Hillman
Rod Bogart

Discussion:

  • Joseph: VES Tech committee is undertaking an investigation of metadata. First step is s survey, to figure out how bad things are (i.e. there are reports of 13 different definitions of focal distances).

  • Kimbal: found a bug in one of the unpackers. Unpacking rgb-only and converting to float has a bug.

  • Kimball: Weta will be producing production images with the new core.

  • https://github.com/AcademySoftwareFoundation/Imath/pull/202 “Cuda safety fixes”

  •  

    • Larry: just trying to reduce warnings. Still sorting out some warnings, but they seem to be just moving from one place to another.

    • What’s in 3.1 is fine. And if fixes a CUDA compile error.

  • Cary: Imath readthedocs build has started failing.

  •  

    • Larry: could be sphinx/breathe version. OIIO specifies the version in requirements.txt.

    • Nick: OpenTimelineIO specifies it in conf.py

  • Imath #195 PyImathTest ends with SIGSEGV on Fedora/RHEL

  •  

    • Cary: I can’t reproduce.

    • Kimball: seems to be a range check error. Kimball will investigate.

  • #1125: RFC: switch Zip compression level from 6 to 4 by default, and add attribute to control it

  •  

    • Could add a build-time configuration. Then change the default later.

    • Continue to use zlib but use deflate.

    • Larry: don’t need a tremendous amount of rigor: a few renders, a few captured images. Should be OK.

    • Larry: OIIO patch: cleaned up timing things. It’s amazing how long a big write takes. Do a bunch of stuff and the write takes 80% of the time. 

    • Should offer run-time controls. If nobody gets to it right away, Larry may investigate.

  • Core library:

  • Can’t code/decode DWA compression.

  • Need to finish up the deep packing.

  • How/when do we approach integrating the core into the existing C++ API?

  •  

    • 4.0?

    • Kimball: I was thinking we need to change the member fields of the classes, and thus ABI change. 

    • Larry: That doesn’t matter. Just recompile your code.

    • Could use thread-local storage. Use a thread id to store which thread the framebuffer 

  • OpenVDB:

  •  

    • We need to take the OpenEXR 3.1 integration into OpenVDB seriously, sooner rather than later. If OpenVDB misses the deadline, the VFX reference platform will revert OpenEXR back to 2.4.

    • Kimball will contact Nick Avramoussis.

    • Cary will contact Dan Bailey.