Working group meeting notes for December 19, 2019
Attendees: Benedikt, Victor, Amos, Mohamed, Mark Agenda Review action items from last meeting: Benedikt - continue working on joining .NET foundation Benoit - multithreaded python code ? - breaking up Quantconnect PR ? - github actions Amos - continue breaking up pull 958 into smaller PRs to make it easier to review and merge. e.g. have a single PR that gets reference counts right and doesn't touch GC chains. See Benoit's example of the __import__ restore PR for best practices. Amos - same for other older PRs Benedikt - will try to reproduce PyBuffer crash locally Benedikt - will check if there are any breaking changes since 2.4.0 Benedikt - continue exploring CLR loader, splitting parts into separate Python interop DLLs, have main Python runtime implement layers on top Victor - Perf runs in CI Review PRs Mailing list currently moderated - @markv asked for ownership to be transferred to Benedikt Notes joining .NET process had changed, Benedikt working on it Pull 958 (domain reload handling) split PRs being approved and merged, thanks Amos! github actions if we join .NET we can stay on current setup would be nice to have Darwin CI working PyBuffer nothing done yet CLR loader modernization PR has a single build Python 2 should still work, but Benedikt won't spend a lot of time on Python 2, 3.5+ is current focus moved code and tests into separate directories removed old csproj file in favour of small project files generated using .NET CLI goal is a fully building project using .NET CLI that is operating system independent made a small pythonnet module that will be the future entry point still rough, testing currently with core CLR and Mono on Linux loading works, P/Invoke __internal breaks on .NET core Python functions have been split out into an interface using T4 templates ready for early review if you can take a look Performance tests and CI not very stable if we put in CI, fail depending on weather conditions no way to put in CI without random failures maybe makes sense to keep it as an ongoing statistic on master and not blocking PRs useful for running locally on branch vs. master to check for regressions still informative for detecting regressions - we can run it on every PR, but it shouldn't result in a hard error - can we report it on the PR? perf comparison is relative, so different machines shouldn't affect the result Mailing list keep it, but remove moderation, transfer ownership to Benedikt Improve method binding PR #974 discussion Meeting on January 2 is cancelled, next meeting will be in 4 weeks on Thursday, January 16, 2020 Action Items Benedikt will continue working on .NET foundation application Victor will take a look at the CLR loader PR, looks big though Victor will look into if we can generate perf stats through AppVeyor and Travis Victor will publish a PR for conversion config Mohamed will split PR #974 Benoit - multithreaded python code ? - breaking up Quantconnect PR Benedikt - will try to reproduce PyBuffer crash locally The meeting notes google doc is here https://docs.google.com/document/d/1rJVU84B_dgx58-_EopjRtOJVFAI2WfHJYV0n7uE1.... Feel free to correct or add additional information. Please note there will be no meeting on January 2nd. The next meeting will be held on Thursday, January 16th at 12pm EST, 9am PST, 6pm CET, 1am China. Happy holidays! Mark Visser Tooling Dev Manager Unity Technologies - www.unity3d.com http://www.unity3d.com/
participants (2)
-
Denis Akhiyarov
-
Mark Visser