TSC Meeting Notes 2021-04-07

Owned by Cary Phillips
Last updated: Nov 04, 2022
2 min read

Attending:

  • Cary Phillips

  • Christina Tempelaar-Lietz

  • Eskil Steenberg

  • Kimball Thurston

  • Larry Gritz

  • Peter Hillman

  • Nick Porcino

Discussion:

  • Issue #996 The exception occurs inside the call: ImfOpenInputFile().

    • Almost certainly something that's been fixed, the stack trace is identical to others. It's from OpenEXR v2.3.

  • Issue #136 Cmake targets provide wrong include dirs and libraries.

    • Nick: Most likely user error, it seems to be doing the right thing.

    • Need to confirm that OpenEXR works properly as a subproject.

    • Larry: I'm experiencing something weird with OIIO. OpenEXRConfig.h is missing, I don’t understand the transitive dependencies. Not sure what’s going on. But I may have some insight to share.

  • Is LibraryDefine.cmake necessary?

    • Kimball: The CMake book says there should be a config pseudo library that’s generating the Config.h.

    • LibraryDefine.cmake can probably be removed. Probably a result of moving everything together in the reorganization.

  • PR #995 Don't impose C++14 on downstream projects

    • Proposing a change. We compile for c++14 by default, but we make it impossible for downstream projects to use C++11.

    • Only affects downstream projects that consume OpenEXR/Imath that wants to use C++11. It was a downstream user of OIIO.

  • CI:

    • Would be useful to have a basic project that consumes OpenEXR, as a sanity check.

  • 3.1/Core

    • Kimball: done a clean up of the API. May make one more change. I’m pretty close, basic writing is finished. Writing tests.

    • What is the sequence for integrating this with the C++ side?

    • Can you mix and match? Yes, they’re totally separate. I have performance tests of reading with each.

    • Want to be able to convert the type from half to float, or have to include the half-to-float table.

    • Is that necessary in the C API? It’s mean to be lightweight. A lot of these does-it-under-the hood thing end up being performance problems.

    • ES: As a C programmer, I’d like that conversion.

    • ffmpeg requires a C api.

    • Nick: Just do this:

      #if __cplusplus__ struct half_t {} #else class half { ...

    • What about fuzz tests? Peter: we have our own fuzz test. Imf::checkFile() is the entry point. C API could be added to that.

  • 3.1 branch?

    • 3.1 would be both API’s but independent of each other.

    • 3.1 -> 3.2 is fine as long as source doesn’t change, but ABI can change.

    • There already is a C API. It’s t2o functions that reads a 4 channel rgb single part. Too simple to be useful.

  • Weta CLA? Will sort it out.

  • VES discussion at TAC meeting:

    • May is the draft of the reference platform.

    • Sunset guidelines? No hard answers. But they’ll go away and consider it.

    • It puts a burden on the CI.

  • Security: how far back should we support?

    • Should announce a policy change, something like "We will support 3.1, 3.0, and 2.5. Backport as requested.

    • How hard is it to backport to 2.4? Things have changed, it might be error prone.

    • New fuzz issues aren't easy to test in 2.4 because that release doesn't include the test program.