Classes don't seem to be always put in `compiledClassesAndSemanticDbFiles.dest`

[kasiaMarek]

Classes don't seem to be always put in `compiledClassesAndSemanticDbFiles.dest`

[lefou]

Could this be some unsharpness in the BSP protocol spec? Metals wants to use some location to access JVM byte-code of a specific module and it's dependencies, but never requests the JVM byte-code to be produced for the dependencies.

This is also a limitation of the BSP protocol. Although, sementicDB files are not needed to debug something, BSP only supports a single output location for target files of a JVM bytecode files and a semanticDB files producer. So, Mill needs to provide the already produced JVM-byte-code in a shared location with the semanticDB files, which is only available after you have requested and produced both.

We could and probably should fix this on Mill's side by letting the copiledClassesAndSemanticbFiles target depend on its companions of its recursive dependencies. But it would be nice to also open an issue for BSP to get that detailed expectation into the written BSP spec.