TSC Meeting Notes 2021-03-25
Attending:
Cary Phillips
Christina Tempelaar-Lietz
Eskill Steenberg
Joseph Goldstone
Kimball Thurston
Larry Gritz
Nick Porcino
Serio Roja
Discussion:
nvcc 10.1 errors on the detector templates in ImathTypeTraits.h. Not sure what's going on, but #if-ing them out is sufficient.
Get final changes in for the 3.0 release over the weekend, then cut a new beta release, say, Monday. Then let it sit unchanged for several days until April 1, when that exact code becomes the release.
We need better coordination with the VES committee regarding release schedules and vendor demands.
ImfAttribute symbol warnings: not propagating a setting? Export is stubbed out for Windows, that may be the issue. Nick will investigate.
Setting the tag on Imath that openexr pulls to build. Once Imath 3.0 is officially released, set the RB-3.0 branch of OpenEXR to pull from that. But leave openexr/master pulling from Imath/master. That way, we have some CI pulling on the master branch from another repo.
CLA status: Weta and Epic still in progress.
"Release” branch: The goal is that the thing called "release" is always the latest. Have a branch called "release" and update it to point to the release tag after a release. It's not a branch to merge into, and doesn't have to be fast-forwarded, just set to the proper commit.
Release notes technique: Larry: I heavily edit.
Windows Deadlocking: NickP says it's the hopper, haven’t forgotten. The person who submitted gave a minimal repo case. It’s on shutdown. Intermittent and only on Windows. Deadlocks on shutdown. Possible solution: dllmain? Some users wish OpenEXR didn't have custom threadpool, just use TBB.
Some users keep long-lived cmake caches. We’ve baked the namespaces into cmake variables but namespaces update on every minor release. Should the default be a empty string, require setting?
Larry: the build/install instructions should be: “Start with 'rm build'” so there's no chance old cmake caches mess up the build for users who aren't cmake experts.
And CMake 3.13 has -S/-B options: cmake -B my/build/dir -S my/src/dir
gh command line tool is your friend. https://cli.github.com/
Joseph: The SMPTE committee working on the 5-year revision of the ACES Container has taken a careful look at its charter and thinks the group should limit itself to color-related metadata changes in their efforts. Back in 2012, when we created the original document, there was no ASWF, OpenEXR development was effectively dead, and so we though we were doing something useful by putting in some VFX-supporting attributes into the ACES Container specification. That utility was never realized because no one moved those attributes into the reference implementation, though.
The PR that I had put in to reconcile the two worlds, implementing in ImfStandardAttributes.{h, cpp} and the accompanying unit test, doesn’t make sense now. I do think there is utility in moving SOME of that stuff into the reference implementation, but I think the semantics and the names and the interrelationships need to be defined by VFX professionals.
Certainly we have a lot of VFX professionals in the OpenEXR TSC, but many of us are constrained by our employers in what we can say about useful lens metadata. I have proposed to the VES, in email, the idea that they do it, or rather, that the executives who form the VES Technical Committee designate people at their companies who have the expertise, and that (crucially) they give them permission to speak up. Not to divulge their most advanced techniques, but to at least give feedback upon what I propose to move over from the ACES container spec. They agreed to hear me out, and I am presenting during their meeting on the 8th, a week and a half from now.