Missing ArgAbi and BoxRet trait implementations for datums

[danbluhmhansen]

After updating to version 0.12.0-beta.0 I can no longer use JsonB as a pg_extern function argument, and I can no longer use JsonB, AnyArray, or AnyElement as state type in pg_aggregate. Unfortunately, pgrx lacked any tests for these types, so they were not checked for compilation when adding ArgAbi and RetAbi.

[eeeebbbbrrrr]

Interesting. Surprised we don’t have tests to catch these.

Thanks @danbluhmhansen. I’ll approve CI and we’ll see about merging next week, I imagine.

[workingjubilee]

This should land with tests that prevent this omission in the future. Just simple "accepts the type, returns the type" somewhere in pgrx-tests is all.

[workingjubilee]

I have a suspicion that these BoxRet impls for AnyArray and AnyElement will prove... optimistic, but if we can at least make it through a function call, then at least it kinda works.

[workingjubilee]

Preferably, the Any* types should at least have the mercy to have their impl check that you created an AnyType with the same OID returned by get_fn_expr_rettype. Perhaps only beneath cfg(debug_assertions)? That can wait for a separate PR if you are not interested in writing it, however.

[danbluhmhansen]

This should land with tests that prevent this omission in the future. Just simple "accepts the type, returns the type" somewhere in pgrx-tests is all.

Alright I have added tests that should catch possible future regressions.

• json argument type tests to json_tests.rs
• anyelement argument type tests to new anyelement_tests.rs
• anynumeric argument type tests to new anynumeric_tests.rs
• aggregate state type tests for types json, jsonb, anyarray, anyelement to aggregate_tests.rs
  • I couldn't figure out how to write an actual pg_test for the anyarray aggregate state so I left that out, besides any regression should be caught by having the pg_aggregate implementation, regardless of the tests.

[workingjubilee]

Thank you.