TSC Meeting Notes 2021-12-02

Attendance:

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

Agenda:

Key Dates for Review:

Discussion:

  • Google Summer of Code ideas:
    • Add support for bfloat16
      • Is adding a new pixel type safe to do? There’s support for adding a new compression type. But it may not be the same for pixel type.
    • Add a part type that is only metadata. 
    • A fast header read
      • So you don’t have to read the file to get to the header data.
    • Add support for sorting of attributes.
      • Peter: I’ve never grasped why people want to change the order.
      • Larry: They don’t. The only thing you get is alphabetical, which isn’t always what you want. 
      • Peter: an ordering attribute that tells the order. They’re written and read in alphabetical order is they’re stored internally in a map that keys off the name.
      • You can write any order you like, but you can’t ask the API what order they’re in.
      • Joseph: If attributes are written in alphabetical order, would identical files checksum to the same value?
      • Kimball: They are, because they’re a map.
      • Peter: But compression data is not the same.
    • Convert boost python to pybind11. 
    • Explore other compression schemes. 
      • Should try Zstandard.
    • Performance metric suite.
    • A new spectral attribute type.
      • Alban’s github repo looks nice.
    • Switch c++ library to use the C core.
      • Kimball:: I started working on ImfHeader. You can blindly copy a Header object
    • Most hated github question: doesn’t build on fedora, i386, yada yada
  • CMAKE_PREFIX_PATH:  This is, in fact, the proper way to direct OpenEXR to an Imath installation.
  • For python cmake issues, we should look at OTIO does.
  • #1201 Extend multipart RgbaInputFile API
    • After merging #1194, there are conflicts. Peter will investigate.
  • Peter: I can’t reproduce some of the OSS-fuzz issues. And some have been closed without any fixes. Something must have changed internally with the fuzzing system.