Static-only frozen dataclasses?

[kkom]

I'd like to enforce immutability in a popular Python GraphQL server library, which uses data class transforms to represent types.

Here's my original issue report and a proposed solution:

• Can Strawberry created dataclasses be frozen by default? strawberry-graphql/strawberry#3396 (which references changes made to Pyright since 1.1.328
• Make all transformed data classes frozen by default strawberry-graphql/strawberry#3397

As you can see in strawberry-graphql/strawberry#3397 (comment), making all dataclasses frozen introduces a non-trivial performance penalty of 30%. I understand this is because frozen dataclasses are more expensive to initialise: https://rednafi.com/python/statically_enforcing_frozen_dataclasses/

Since immutability is so convenient for static type checkers, I'm wondering if data classes in Python could be extended with a zero-overhead, static-type-checker-only immutability config option?

PS: This is a general Python static typing musing, not really related to Pyright. I'm posting it here because I'd love to hear @erictraut's views on this – but if there is a better place for something like this, please let me know and I'll use it going forward.

PPS: For reference, here is an analogous hack for those using the attrs library https://threeofwands.com/attra-iv-zero-overhead-frozen-attrs-classes/ Something like this would be perfectly fine for codebases that have good static type checking coverage.

[erictraut]

A change like this would need to be made to the CPython runtime or the Python typing spec. The Python typing forum would be a good place to post.

[kkom]

Thank you Eric – I've created https://discuss.python.org/t/static-only-frozen-data-classes-or-other-ways-to-avoid-runtime-overhead/46968 following your advice.