TSC Meeting Notes 2020-07-16

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

  • Rod Bogart

Discussion:

  • The CLA. Disney refuses to sign the current one, although Disney employees already authorized have apparently somehow slipped through. This came to a head with OSL. We (OpenEXR TSC) have no concerns about moving to the new CLA that the ASWF Governing Board legal committee settled on, which is Apache 2. We will request help from John Mertic to navigate through changing the forms in EasyCLA, and getting respetive companies to sign the new one, without anyone getting locked out of push access to the repo. Also, we need to update the documents in repo and make sure there are no references to the outdated form. Cary will request help from John Mertic.

  • Review the logo concepts from Ali Rowan, graphic designer with the Linux Foundation. The red version is nicely suggestive of the old logo. The trapezoidal shapes were intended to suggest a perspective projection, although the shape in the concept designs provided is not a perspective, and this feature has relatively minor appeal to the group. Therefore, we settle on the square-ish red logo. Note that we need several variations: text on the bottom, text to the right, no text, gray-scale, very small "favicon". Cary will request that of Ali.

  • Open Source Day

    • Outline:

      • Brief recap of the year

      • Overview of OpenEXR 3.0 release

      • Overview of Imath (given by Owen)

      • Introduction of new openexr.com (hopefully given by Sarah Martin)

    • Arkell agreed to provide some updated images for the openexr-images repo, although he reports that he's not prepared to generate and deep exr images.

    • Cary requested 25 minutes, but given that Imath is a major topic, we expect enough comments and feedback that we should increase that to 55 minutes. Cary will follow up with Emily Olin.

    • Larry notes: This is the biggest release in a very long time, with signficant implications to the user community. Everyone who builds openexr is going to be affected.

    • We should expect lots of questions and vigorous discussion. We had good questions last year (e.g. “Is OpenEXR appropriate for ML?”)

    • The title of session should prominently feature "Imath".

  • 3.0 Release: time to get in any changes that break compatibility.

    We should make a pass at cleaning up exceptions in the OpenEXR tree.

    Some Issues for potential inclusion:

    • getChunkOffsetTableSize #740

    • Include guards #636

    • Signed strides: #614

    We will need to take care with how OpenEXR references Imath. It could potentially be an external project path in cmake that auto-clones imath repo. Nick reports that this works well with OpenTimelineIO.

    We should remove legacy thread support, just use std::thread We still want to provide the semaphore class. Lock is slower than a semaphore class. Keep the shim layer.

    We should we move some of the directories up a level, so the OpenEXR repo has subfolders:

    • src

    • test

    • tools

  • OSS-Fuzz: Going well so far, around a dozen issues, which have been easy to reproduce and identify. It takes some thought to determine how to resolve them. Do we want to break compatibility, or suppress it?

    Peter points out that the most recent report, the test case is broken, the exr is bad data. Peter will follow up with fuzz folks.

  • Owen reports on progress with Imath.

    • Initial push of constexpr changes

    • A small set of tests. There are some python binding missing.

    • Cuda: The strategy is when compiling with Cuda, put half into Imath internal namespace.

      What is different between cuda and imath half? Bit layout is the same. Our class has HasInf, could use std::isinf override. Rod: We tried to merge with NVIDIA.

      Owen will write a test that verifies the behavior. Should still support C++11.

    • Create a RC3 branch.

  • clang-format: Would like to reformat before the 3.0 release. Nick says that OpenTimelineIO used the openexr template with success. Larry notes: Whatever we choose, there will be sections that don't format nicely, so just turn off formatting. But, a year later, you'll forget about what you didn't like.

    Don't like #include sorting, it can break the build; alphabetical order is not always what you want.