TSC Meeting Notes 2020-07-30

Owned by Cary Phillips
Nov 04, 2022
3 min read

Attending:

  • Cary Phillips

  • Christina Tempelaar-Lietz

  • Joseph Goldstone

  • Kimball Thurston

  • Larry Gritz

  • Nick Porcino

  • Owen Thompson

  • Peter Hillman

Discussion:

  • Joseph proposed discussing the pros and cons of implementing the attributes in the ACES Container spec (SMPTE ST 2065-4) in the reference implementation. Putting them in the reference implementation will effectively make them more portable. OTOH any organization using unqualified names (not ‘legal’ according to the spec, but of course the reference implementation doesn’t stop you from doing it) in a nonstandard way potentially could see an impact.

    • Joseph: I have a laundry list of things I’d like to include. The ACES container has never been a subset of OpenEXR. People who want to use the standard attributes have to implement them themselves.

    • If you have two different companies and they’re each implementing fstop, one does it as a string, one as a float. If it’s in the reference implementation, it would force consistency. You can read a raw file and find some things in exif fields, other things from somewhere else. What happens if the two things have the same name? What happens if the same thing is in two different places?

    • Has anyone added standard attributes recently? No one has done that in a very long time. Gitlog says nothing has been added since 2014.

    • Joseph: Please look at the list of standard attributes and report which ones are used at your shop.

    • Larry: A lot of these correspond to things we see all the time. A lot of these, because they correspond to XMP, etc, by convention, the OIIO does say “It’s supposed to be called blah.”

    • But there's a difference between standard attributes and required attributes. This list combines required and standard.

    • Nick: I propose to adopt the SMPTE names. Move forward with SMPTE standard.

    • Larry: we should really strongly recommend that if anybody does anything non-canonical, they should prefix it.

    • Peter: It’s not the file format’s job to interpret the attributes. But if we can provide guidance and helpers to applications.

    • What’s in the standard attributes file suggests what the type of an attribute is, but you don’t have to use it. You can insert any attribute with any type. The library doesn’t enforce the attribute usage.

  • Update on Imath

    • Most significant contribution: restructuring of all the source files. It’s been pushed.

    • Installation of imath: tried to do external project. Kimball: we can follow up.

    • Nick: Can you add zlib, too?

    • Would be nice if it fails to download.

    • Nick: remember that OpenTimelineIO has a system that’s working with our CI.

    • What should the version number of the Imath release be? 1.0? That might conflict with the libImath from OpenEXR 1.0. It's probably best to start with "3.0", odd as that sounds. Then OpenEXR and Imath can diverge from there as necessary.

    • Nick: Radical suggestion: D&I initiative: rename it to main. Rebasing on master is gross. Could change to "main", "dev".

    • Larry: This is being hotly debated. Let's wait for a consensus to emerge in the community about what everyone si going. Wait to jump on the bandwagon.

    • Owen't next step: Finish up the pull request, then get back to CUDA.

    • Larry: two more things before the release:

      • Cuda decorators: make a define if you’re being compiled by the CUDA compiler, define host device. Then decorate all functions with that.

      • Change the individual component storage to a union, which provides performance improvements.

      • Owen: I'm planning to do both of these right away.

    • Should also replace std::isinf for half.

    • Maybe an option to put half in a namespace?

  • We should aggregate all the “can’t find boost” issues into one, since they're all versions of the same problem. With the most recent boost release, cmake errors: "Boost is incorrectly configured.” Kitware and boost in a fist fight.

  • Still getting permission issues on sonar scan stuff.

  • Reviewed overview of slides for ASWF Open Source Day, scheduled for Wednesday, August 19, 4:30pm PT.

  • #636 weak namespace and include guards.

    The modern idiom is to use #pragma once. Header guards prevent including twice, but also allows you to know whether they’ve been included. We should not take them out, they're still useful. You can get to the same file in two different ways.

  • Other issues not discussed:

    • #789 Better stream interface for DeepScanlineInputFile

    • #773 PyIlmBase cmake is able to find Boost but not boost::python

    • #750 Zero allocated memory to guarantee always identical output in DWAA compression

    • #740 Remove deprecated argument to getChunkOffsetTableSize

    • #548 mark half(float) constructor as explicit

    • #485 Remove _maxScanLineSize members