Add definitions for Postgres' bytea type

[ngld]

I was surprised that pggen doesn't support bytea columns. Was that intentional? I've added the necessary type definitions and successfully used the result.

I've also tried to use --go-type 'bytea=github.com/jackc/pgtype.Bytea' but that doesn't work since pggen still complains about the missing OID: https://github.com/jschaf/pggen/blob/7f6f84eb1636947992128675bddf9b33c5e2d557/internal/pg/types.go#L355-L361

Did I misunderstand how --go-type is supposed to work or does it only allow remapping known OIDs to different Go types?

[jschaf]

Thank for the report and commit. I think there's 2 bugs here:

1. pggen errors too early when it can't determine a type for an OID. The code you highlighted throws an error if a type is not an enum type, a composite type, or an already known type. Throwing an error is a bug because we might be able to match the type by name using the --go-type flag.

2. No support for the known type bytea. Your PR fixes this bug. I copied all the OIDs from https://github.com/jackc/pgtype/blob/1e141d8c32939b0c0fb2fac854cd37fc543b7835/pgtype.go#L17 and used a SQL query to generate the first version of the mappings (since changed by regex manipulations). I'm not sure how I missed bytea; the OID is in the sql query (17):

WITH types AS (
  SELECT oid,
         typname,
         replace(
             replace(
                 replace(
                     replace(
                         replace(
                             replace(
                                 replace(replace(replace(replace(replace(
                                                                     replace(regexp_replace(typname, '_(.*)', '\1Array'), 'oid', 'OID'),
                                                                     'cidr', 'CIDR'),
                                                                 'bpchar', 'BPChar'),
                                                         'aclitem', 'ACLItem'),
                                                 'uuid', 'UUID'),
                                         'jsonb', 'JSONB'),
                                 'json', 'JSON'), 'char', 'QChar'), 'cid', 'CID'), 'xid', 'XID'), 'tid', 'TID'), 'acl',
             'ACL') AS cap_name
  FROM pg_type
  WHERE oid IN (
                16, 17, 18, 19, 20, 21, 23, 25, 26, 27, 28, 29, 114, 600, 601, 602,
                603, 604, 628, 650, 651, 700, 701, 718, 705, 829, 869, 1000, 1005,
                1007, 1009, 1001, 1014, 1015, 1016, 1021, 1022, 1033, 1034, 1041, 1042,
                1043, 1082, 1083, 1114, 1115, 1182, 1184, 1185, 1186, 1231, 1560,
                1562, 1700, 2249, 2950, 2951, 3802, 3807, 3912, 3904, 3906, 3908, 3910,
                3926
    )),
     go_types AS (
       SELECT oid, typname, upper(left(cap_name, 1)) || right(cap_name, -1) AS upper_name
       FROM types),
     decls AS (
       SELECT oid, typname AS pg_name, upper_name AS go_name, upper_name || 'OID' AS go_type
       FROM go_types
     )
SELECT string_agg('  ' || go_name || ' = Type{OID: pgtype.' || go_type || ', Name: "' || pg_name || '"}', E'\n')
FROM decls;

I'll merge your fix, add a test, and then fix the first bug.

[jschaf]

The first bug is fixed in 5c4f88a. I created a new pg.Type: UnknownType to delay unknown OID errors until the Go code generator has a chance to choose a mapping based on the name of the type from the --go-type flag.