From bae142474548d36737324e5c34f3db73d3eaf561 Mon Sep 17 00:00:00 2001 From: Dan Bailey Date: Thu, 3 Nov 2022 10:20:45 -0700 Subject: [PATCH] Add meeting notes for 18 Oct Signed-off-by: Dan Bailey --- tsc/meetings/2022-10-18.md | 88 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 88 insertions(+) create mode 100644 tsc/meetings/2022-10-18.md diff --git a/tsc/meetings/2022-10-18.md b/tsc/meetings/2022-10-18.md new file mode 100644 index 0000000000..557fac67a5 --- /dev/null +++ b/tsc/meetings/2022-10-18.md @@ -0,0 +1,88 @@ +Minutes from 151st OpenVDB TSC meeting, October 18th, 2022 + +Attendees: *Jeff* L., *Andre* P, *Dan* B., *Rich* J., *Ken* M., *Nick* A. + +Additional Attendees: Greg Hurst (United Therapeutics), Sebastian Gaida, Barry Dempsey + +Agenda: + +1) Confirm quorum +2) Secretary +3) VDB Mathematica Link +4) AX +5) Houdini LZMA +6) ValueAccessor Performance Improvements +7) CMake VDB Tool +8) Next Release +9) Next meeting + +------------ + +1) Confirm quorum + +Quorum is present. + +2) Secretary + +Secretary is Dan Bailey. + +3) VDB Mathematica Link + +Almost ready for this PR to be merged. Documentation still to be finialized. This plugin will not be +built through CI, still waiting for progress there. Question about licensing. One of the 3rd party +dependencies is LTemplate which uses a different license to VDB. MIT License is compatible and does +not do copy left so no further action needed. + +4) AX + +Nick has looked at LLVM 15, but significant work is needed to support this so this version will +explicitly not be supported in VDB 10.0. + +Weekly CI is still failing with a Blosc 1.17 error, if all checks are green seems that we can +disable this error now. + +In LLVM 14, a new FMA option which was previously off by default has now been enabled. This is done +at the IR level so will be hard to work around. This is exposed using the fpcontract compiler +flag. + +5) Houdini LZMA + +Jeff has looked into the LZMA issue in Houdini and confirms the problem that CentOS7 now uses a +different patch version of LZMA to the one that Houdini ships with as a result of the dependency on +boost::iostreams. There are some options, including to extend the LZMA shipped with Houdini to +include the new symbols from the updated CentOS7 version. This issue will go away as users +transition away from CentOS7. + +For the CI where this is a problem, we can circumvent this issue by disabling delayed loading which +means boost::iostreams is not used. This is the route we will go for the VDB 10 release to switch +back on the Houdini plugin builds. + +6) ValueAccessor Performance Improvements + +In addition to the performance improvements, this PR fixes a couple of bugs with adding leaf nodes +through the ValueAccessor. Jeff highlights that some of these changes were proposed in PR1282. Nick +to look over this PR and compare his implementation. One component included in this old PR which is +not present in the new one is the change to fix for adding tiles through the ValueAccessor. + +7) CMake VDB Tool + +The CMake build for the vdb_tool uses inconsistent names for the CMake variables and does not use +targets for compile options and definitions which risks creating confusing behavior when switching +on this tool as these options will apply to all the subsequent components. Andre to look into +addressing both of these issues before VDB10. + +8) Next Release + +Continue to look at resolving all the Weekly CI issues and the VDB10 release will be started next +week. + +There is a hesitancy to do more frequent releases because the release process is perceived to be +slow. This is something the TSC should re-evaluate based on the ease of doing the VDB10 release. +Only two releases in the 9.x series. Ideally, the next minor release should follow in under six +months. + +Dan commits to look at the PointReplicate PR soon after the VDB 10 release. + +9) Next meeting + +Next meeting is on October 25th, 2022. 2pm-3pm EDT (GMT-4).