Skip to content

2023 02 28 meeting

Artem Pelenitsyn edited this page Mar 1, 2023 · 1 revision

Meeting 2023-02-28: Types Database

Insight: Incompleteness of Enumeration Can Be Improved Via Types From Tests

If we can’t enumerate everything starting from top, we could try some types from the bottom. Which types exactly? The types seen in tests from many Julia packages.

Likely Types

Tests will have a lot of types — which ones to take? We can introduce an abstract notion of likely types. First approximation for implementing it would be to take more frequent types. But perhaps we could go further at some point and build an ML model that could spit some types that should be more appropriate for our situation.

How To Benchmark? Compare to tests

Tests will hit some types. Enumeration will either go through all possible (in that case we definitely better than tests), or it will only try likely types, in which case we can see if we managed to at least hit the types that the developer deemed important (implicitly).