hasql: array-of-domain columns need a name-resolved codec, and fail only at run time

[alexbiehl]

Symptom

A query selecting a column whose type is an array of a domain fails at run time with

UnexpectedColumnTypeStatementError 33 1001 211356
                                      ^     ^ the domain's array OID in this database
                                      bytea[]

Nothing catches it at compile time. It surfaces the first time that query runs.

Why

hasql validates each result column's OID against the decoder. For a scalar domain column PostgreSQL reports the domain's base type in the row description, so Hasql.Decoders.bytea decodes a bytea-backed domain column fine — which is what sqlc-hs emits, and it is correct. For an array of a domain it reports the domain's own array OID instead, and the plain bytea/text decoder's array OID no longer matches.

The array OID is assigned per database, so it cannot be a constant.

What the fix looks like

Hasql.{Encoders,Decoders}.custom takes a type name and Nothing for the static OIDs, and hasql then resolves both the scalar and array OID from pg_type by name at run time. So for an array-of-domain column the codec wants to be

Hasql.Decoders.custom Nothing "opcua_attribute_value" Nothing [] (\_lookup bytes -> ...)

rather than Hasql.Decoders.bytea. Encoders.custom takes one argument more than the decoder — a render function used only in error messages.

sqlc-hs already has everything it needs to decide this: column ^. #isArray says it is an array, and columnDataType (column ^. #type') is the domain name. hasqlColumnCodec is where the choice would go.

Workaround today

Per-column hasql_encoder / hasql_decoder overrides naming a hand-written codec, one pair per column, plus nullable: true and nullable: false variants of each (see the other issue about nullable defaulting). In an application with four such columns that is eight override entries and two hand-written codec pairs to express something the plugin could derive.

Finding the affected columns in a schema:

SELECT c.relname, a.attname
FROM pg_class c
JOIN pg_attribute a ON a.attrelid = c.oid AND a.attnum > 0 AND NOT a.attisdropped
JOIN pg_type t ON t.oid = a.atttypid
JOIN pg_type el ON el.oid = t.typelem
WHERE c.relkind = 'r' AND t.typcategory = 'A' AND el.typtype = 'd';