Normalize setter sig param names from exported gem RBIs

[dduugg]

Motivation

Some gems (notably stripe-ruby) ship exported RBIs where setter sigs use the attribute name with a leading underscore (_expand, _payment_intent, etc.) as the parameter name. However, stripe overrides attr_accessor with define_method("#{name}=") { |value| ... }, so the runtime setter parameter is named value.

After tapioca's Keep::LEFT merge, the method def comes from runtime introspection (value) but the sig is pulled from the exported RBI (_expand). Sorbet rejects this sig/def mismatch with error 5003, producing type errors in codebases that depend on the stripe gem.

Implementation

After merge_trees, the merged tree is walked by a new SetterSigParamNormalizer visitor. For each setter method (name ends with =) with a single ReqParam, any attached sig param whose name differs from the method def's param name is replaced with a new SigParam using the method def's param name. The type annotation from the exported RBI is preserved — only the parameter name is normalized.

SetterSigParamNormalizer is an RBI::Visitor subclass defined inside Commands::AbstractGem and marked private_constant.

Tests

Added an integration test that creates a mock gem with a define_method-based setter (runtime parameter value) and a bundled RBI whose sig uses _expand, then asserts the generated RBI has params(value: ...) in the sig.

[KaanOzkan]

Looks to be same as #2589 which should be fixed in rbi. I'll release a new version.

EDIT: Done, try v0.3.11 for the rbi gem

[dduugg]

Thanks, @KaanOzkan !