Record diagnostics source rule when testing

[pepeiborra]

I have seen unexpected duplicate diagnostics leading to a failure in the ghcide testsuite: what seems to happen here is that both Typecheck and GetModIfaceFromDisk contribute the same diagnostics. But this should never happen, because we never run both these rules for the same file:

• GetModIfaceFromDisk is only run for non FOIs via GetModIface
• Typecheck is only run for FOIs, either via GetModIface or on demand by code actions

This PR adds more testing-only logging to help diagnostics when this happens again next time

[fendor]

Nice addition

[michaelpj]

Slightly an abuse of the source field but I guess it's fine. We could put it in the data field? which is entirely implementation-defined. Then we could consider even doing it unconditionally, or having some richer-structured "provenance" structure.

[pepeiborra]

Slightly an abuse of the source field but I guess it's fine. We could put it in the data field? which is entirely implementation-defined. Then we could consider even doing it unconditionally, or having some richer-structured "provenance" structure.

Do you mean the code field? I like the idea, but it doesn't look like diagnostics have a data field

[michaelpj]

No, code is for the error code, which we might actually get from GHC before long. Diagnostics do have a data field, but not in the version of the lsp library we're using, I guess.