Running Test and Perf Improvers: Status

[dsyme]

@kMutagene Linking this issue from FSharp.Stats as the offer applies across all of fslaborg/

fslaborg/FSharp.Stats#353

[kMutagene]

@dsyme i think this is a very good candidate repo to test this without any blockers. Any steps we need to take?

[dsyme]

@kMutagene Please add comments on these

• Test Coverage Improver Plan
• Perf Improver Plan

There are also a couple of minor PRs made by them, nothing much yet. I've disabled them while I iron out a few bugs (we swtiched these actions to use discussions not issues and that's caused a few problems)

[dsyme]

@kMutagene I ironed out a few kinks (hopefully!) and have now set both improvers running again

[dsyme]

@kMutagene I think we have enough PRs to take a first look at the crop

I had to run a few more than I thought becaise still ironing out a problem where the workflows should be adding comments to the discussion about the work they've been doing. I'll do one last run to see if it works

[dsyme]

@kMutagene I think we should start to merge the test additions - they are solid

I will check if the test times are getting too long.

[dsyme]

@kMutagene OK I went through and merged the benchmark additions and tests.

Coverage started around 4.8% - hopefully it's getting over 10% now with all those in.

The proposed perf improvements need much more careful scrutiny and I will leave those open.

[dsyme]

@kMutagene I added this comment to tell the coding agents how to get coverage for inline functions

It results in test duplication however. Maybe it doesn't matter. Clearly from the PRs testing for this library can be basically be fully automated. I think we should just let it run to 100% coverage or as close as it gets.

#5 (comment)

[dsyme]

@kMutagene - OK, so reported coverage has been increased from about 4% to 76%. I will continue to run for a while and see if it continues to increase. There's usually an upper limit to these things. In this case, the inline methods involving Span or Byref are not possible to cover by any technique

On the whole the quality of the library is excellent - the fact there are no detected bugs in SVD etc very promising.

When you get a chance could you take a review and let mw know what you think. We can delete off any tests that you think aren't worth it or will be a significant long term burden for future design changes you plan.

We can also add ongoing coverage reporting in CI/CD. In fact I'll do that now.

On Perf Improvement - the AI tools have less to say, beyond the few PRs. It's not like any huge red flags. The tools are currently saying "we need clarity on the PRs already created". We could unblock that by dropping a comment on the Perf plan in the Discussions. But it's probably a good time to have a think what the overall perf goals should be.

[dsyme]

@muehlhaus @kMutagene Don't forget to close/reopen the "Daily" pull requests as they come in to trigger CI.

It's annoying and I'm trying to get GitHub to change this