One point of having interface files separate is tools like `make`. If the interface file's timestamp did not change, the interface is the same.
It would probably be nicer to have two compilation artifacts, the extracted human-readable interface part and the object-file implementation part.
It (in my case TopSpeed Modula-2) was waaaay faster than e.g. Borland C++ wading through endless .h files.
And how would they test/run their code, since the interfaces would at best link, but not run anything, being only stubs?
>It would probably be nicer to have two compilation artifacts, the extracted human-readable interface part and the object-file implementation part.
Yes, if interface-only coding was possible for the clients, and disk space was at a premium yes. Anything but having to manually write interface AND implementation.
Oberon implementations traditionally generates symbol files, that while I believe they are usually binary tends to have tools to generate textual representations, and there are plenty of tools to generate versions of the modules with implementations stripped as well. To compile a module you should need only the symbol files of the modules you import, not the compiled modules themselves.
Proper modules are great, but I wouldn't call the "separate files for interface and implementation parts" a good feature. That's a throwback to languages like C.
It's not DRY, and it's also busywork -- the interface can trivially be created/inferred from the implementation file alone.