#14960 Codegen for JVM languages does not support inference Issue created by stuhood To declare that they expose JVM symbols, the various codegen implementations which implement JVM code should implement a new

@union

similar to pants/src/python/pants/jvm/dependency_inference/symbol_mapper.py Lines 66 to 81 in </pantsbuild/pants/commit/ed8366859205e071e25f7789d1448f032228b30f|ed83668> , but which exposes package level matches similar to pants/src/python/pants/jvm/dependency_inference/artifact_mapper.py Lines 147 to 149 in </pantsbuild/pants/commit/ed8366859205e071e25f7789d1448f032228b30f|ed83668> Probably the cleanest way to do this would be to merge the (firstparty)

SymbolMap

and

ThirdPartyPackageToArtifactMapping

into the trie structure used by the latter, and to expand the latter to support both wildcard matches and exact symbol matches. That way, inference implementations for the JVM can choose the granularity at which they operate: for example, if an inference implementation doesn't support more accuracy than "this package will be generated", but if they do actually know the precise symbol that will be generated by codegen, they can expose that instead. pantsbuild/pants