Regenerate constitution golden tests #6785
Conversation
ana-pantilie
temporarily deployed
to
github-pages
GitHub Actions
Inactive
ana-pantilie
added
the
No Changelog Required
They take too long, there should be a nightly job which runs them but I don't know which one it is and why it isn't failing. Or maybe it is failing and nobody is being notified of the failures. |
Yes they take too long because all tests (unit,random,golden) are bundled together. |
|
Sounds good @bezirg please open an issue |
| \(z : r) (f : a -> list a -> r) (xs : list a) -> | ||
| chooseList | ||
| {a} | ||
| {all dead. r} | ||
| xs | ||
| (/\dead -> z) | ||
| (/\dead -> f (headList {a} xs) (tailList {a} xs)) | ||
| {r}) |
There was a problem hiding this comment.
This is the old way of doing pattern matching, so performance changes are arbitrary. It looks like something has become lazier and something has become stricter, so it's just all noise. Instead of investigating this we should move to pattern matching builtins and then look for optimization opportunities in generated code.
ana-pantilie
temporarily deployed
to
github-pages
GitHub Actions
Inactive
These are the tests which have been broken on master for a while now. They are slightly less efficient now. We need to update them. AFAIK the pattern matching builtin stuff is what changed the performance here. I don't know how we can fix this performance regression, but until we figure out a way (or maybe @effectfully already has something in mind) we can't just leave these tests to silently fail on master.