Codemods to automatically upgrade for newer Hypothesis versions

[Zac-HD]

Hypothesis is a mature project with a stable API - but we do occasionally deprecate and eventually remove things. It would be nice if we shipped tools to automatically refactor affected code, and LibCST looks great for that!

Codemod ideas

Automatically fix deprecated code, to make updating Hypothesis easier:

• Converting positional to keyword arguments (largest impact and therefore highest priority) (in First codemods: fix positional arguments for keyword-only parameters #2712)
• Converting min_magnitude=None to min_magnitude=0 (in First codemods: fix positional arguments for keyword-only parameters #2712)
• Updating provisional IP address string strategies to the new form
• There were many removals in 5.0.0 which could be codemodded to support slow-updating teams
• If you're still using 3.x and codemods for removals in 4.0.0 would help you update, please let us know

And of course deprecations aren't the only possible use for a refactoring tool:

• Convert assume() calls on a single argument into .filter() calls on the relevant strategy
• Use logic like Return efficient strategies for simple filter predicates, instead of rejection sampling #2701 to suggest converting strategies from filtering to valid-by-construction
• Discard the @st.composite decorator for functions which could just return a strategy, because they have no dependencies between draws.
• other ideas TBD, suggestions welcome

General notes

• I'd like to support this via both the LibCST and the Hypothesis CLI, for maximum flexibility and visibility
• We should ensure that deprecation messages suggest using the codemod if one is available, and add that to our development guide (along with "please add a codemod, or issue, if deprecating something")

[sobolevn]

Important note: libcst does not work with python3.10+ and can break on some python3.9 code.
Related Instagram/LibCST#285

[Zac-HD]

I hope that they'll fix that soon!

In the meantime, I'm optimistic that this will be fairly low-impact - if you're running codemods, that implies you've been using Hypothesis for a while and can probably upgrade library versions before trying the Python version upgrade. Even if some of your test files no longer compile under Python 3.9, using 3.8 to run hypothesis codemod ... will at least handle the others.

[Zac-HD]

Closing this because the remaining ideas don't seem worth the implementation effort; though I'd still be delighted to accept PRs if anyone would like to build more codemods 🙂