awesome! i think there are enough existing tests for correctness, but could you add a benchmark similar to this one using that activerecord query: https://github.com/cockroachdb/cockroach/pull/59880/files#diff-389fae48c6f52a8edcfdb2f3ef5780fb9e349bfad03480d6271470977a24f2d0

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @jordanlewis)

pkg/sql/sem/tree/casts.go, line 1279 at r1 (raw file):

			case oid.T_regtype:
				// Mapping an oid to a regtype is easy: we have a hardcoded map.
				typ, ok := types.OidToType[oid.Oid(v.DInt)]

i guess this is separate from your PR, but does this work for user-defined enum types?

pkg/sql/sem/tree/casts.go, line 1291 at r1 (raw file):

				// from integer to procedure name.
				// Mapping an oid to a regtype is easy: we have a hardcoded map.
				name, ok := OidToBuiltinName[int(v.DInt)]

just curious, is there a way we can warn ourselves in the future if we do support user-defined functions?

pkg/sql/sem/tree/casts.go, line 1299 at r1 (raw file):

				return ret, nil

			default:

is the default case still used anywhere? it would be nice to remove the queryOid function altogether if not!

pkg/sql/sem/tree/casts.go, line 1315 at r1 (raw file):

			case oid.T_oid:
				return &DOid{semanticType: t, DInt: i}, nil
			default:

do we need to look out for this case too?

pkg/sql/sem/tree/function_definition.go, line 165 at r1 (raw file):

// to their name. We populate this from the pg_catalog.go file in the sql
// package because of dependency issues: we can't use oidHasher from this file.
var OidToBuiltinName map[int]string

nit: could it be map[uint32]string ?

pkg/sql/sem/tree/overload.go, line 91 at r1 (raw file):

	// Oid is the cached oidHasher.BuiltinOid result for this Overload. It's
	// populated at init-time.
	Oid DInt

nit: shouldn't the type here be DOid?

i think we want to move in a direction where we are more consistent in the code about the fact that OIDs are always uint32s, so i hesitate to make this a DInt

Added a test, but not sure how to adjust the expectations file (and that test seems to be skipped too): #57771

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @jordanlewis and @rafiss)

pkg/sql/sem/tree/casts.go, line 1279 at r1 (raw file):

Apparently not. Definitely something that needs to be fixed! Filed #61216

pkg/sql/sem/tree/casts.go, line 1291 at r1 (raw file):

Hmm............... I can't think of one. I added it to the tracking issue though. #58356

pkg/sql/sem/tree/casts.go, line 1299 at r1 (raw file):

It is for ::regnamespace. I was thinking about getting rid of this, but we're not ready yet. queryOid is also still used for several other cases.

pkg/sql/sem/tree/casts.go, line 1315 at r1 (raw file):

Good point! We had some special cases above that were not being done here properly. I've refactored a function out that can be used in both places.

I figured out how to adjust the expectations file actually!

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @jordanlewis and @rafiss)

amazing!! thank you for taking this on

left optional comments about typing choices.

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @jordanlewis and @rafiss)

pkg/sql/sem/tree/function_definition.go, line 165 at r2 (raw file):

// to their name. We populate this from the pg_catalog.go file in the sql
// package because of dependency issues: we can't use oidHasher from this file.
var OidToBuiltinName map[uint32]string

err i realize i was the one who suggested making the key be uint32, but perhaps it should actually be a oid.Oid (same as the existing OidToType map)

pkg/sql/sem/tree/overload.go, line 91 at r2 (raw file):

	// Oid is the cached oidHasher.BuiltinOid result for this Overload. It's
	// populated at init-time.
	Oid uint32

and here, perhaps this should either be a oid.Oid or a DOid?

i'll leave this up to you

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @jordanlewis and @rafiss)

pkg/sql/sem/tree/casts.go, line 1274 at r2 (raw file):

		switch v := d.(type) {
		case *DOid:
			return performIntToOidCast(ctx, t, v.DInt)

oh hmm it seems like something has tripped up the oid::regclass cast in the pgoidtype logic test

lgtm!! thank you