pantsbuild / pants / 35131111084

93%

push

github

jvm: resolve protobuf codegen ambiguity per requesting language (#23668)

## Summary

Implements the design proposed in #23648 ("preferred codegen" dispatch)
to fix a bug where a single, unparametrized `protobuf_sources` target
could not be depended on by both a `java_sources` and a `scala_sources`
target at the same time.

When both the Java and Scala protobuf codegen backends
(`pants.backend.codegen.protobuf.java` and
`pants.backend.codegen.protobuf.scala`) are active, this previously
raised an ambiguity error in two independent places:

- JVM classpath resolution (`ClasspathEntryRequestFactory` in
`pants.jvm.compile`) found both `CompileJavaSourceRequest` and
`CompileScalaSourceRequest` compatible with the same `protobuf_sources`
component and raised `ClasspathSourceAmbiguity`.
- Source hydration (`hydrate_sources` in `pants.engine.internals.graph`)
found two matching `GenerateSourcesRequest` candidates for the same
member and raised `AmbiguousCodegenImplementationsException`, once the
classpath-level ambiguity above was resolved and
`compile_scala_source`'s mixed `for_sources_types=(ScalaSourceField,
JavaSourceField)` request reached it.

## Fix

Both are resolved per dependency edge, based on which language is
actually asking — with **no BUILD file changes required**:

- `ClasspathEntryRequestFactory.for_targets`/`classify_impl` take a
`preferred_impl` parameter (the requesting component's own
`ClasspathEntryRequest` type), and `classpath_dependency_requests`
passes `type(request.request)` as that argument. When a codegen input is
shared by more than one impl, only the preferred impl claims it.
- `hydrate_sources` narrows an otherwise-ambiguous set of
`GenerateSourcesRequest` candidates using `for_sources_types` order: the
first `for_sources_types` entry with exactly one matching generator
wins, matching the existing type-compatibility-preference convention of
that tuple. It still raises if no single preference resolves the tie.

51 of 52 new or added lines in 3 files covered. (98.08%)

95884 of 103194 relevant lines covered (92.92%)

3.66 hits per line