Provide a generic encoding escape hatch

[robx]

As an alternative to #146, this provides a way for users to provide arbitrary custom value encoders, via

-- Generic binary encoder, given OIDs and an encoding function.
value :: Show a => LibPQ.Oid -> LibPQ.Oid -> (a -> Binary.Encoding) -> Value a
value oid arrayOid encode = ...

Concretely, this addresses the desire within PostgREST to encode JSON directly from a lazy bytestring. The use would be to replace

SQL.encoderAndParam (HE.nullable HE.jsonBytes) (LBS.toStrict <$> body)

with

let jsonBytesLazy = HE.Value (LibPQ.Oid 114) (LibPQ.Oid 199) Binary.bytea_lazy in
  SQL.encoderAndParam (HE.nullable (HE.value jsonBytesLazy) body

to avoid a copy.

Supposing the general approach seems acceptable, some things I'm unsure about:

• There's the downside that this exposes implementation details to some extent. Should it be exposed as something like Hasql.Encoders.Internal so that it's clear there should be no expectation of stability?
• I'm not at all sure about the naming; also I'd want to document this better (once I know how it ends up).
• To keep things safe in the face of arrays, I chose to make the array OID required, and to fix the format to Binary. Both of these could be relaxed, effectively exposing unsafePTIWithShow.
• Is it a good idea to specify OIDs as LibPQ.Oid? I went with it because it's already a public type...

[robx]

@nikita-volkov any thoughts on this one?

Alternatives might be

• an explicit jsonBytesLazy in the library
• the approach you outlined at https://github.com/nikita-volkov/bytestring-strict-builder/pull/11#issuecomment-1158523706, of giving PostgreSQL.Binary.Encoding a direct path for single strict byte strings; I'd be happy to give that a shot if you prefer the approach

[nikita-volkov]

Ooops. Sorry for dragging on this. I'll get back to you in the coming days

[nikita-volkov]

Please check out the commit above. It adds lazy variations of json functions. If all is fine, I'll make a release.

[robx]

Please check out the commit above. It adds lazy variations of json functions. If all is fine, I'll make a release.

It looks perfect!

(I'm in the process of updating the corresponding PostgREST PR with the required dependency updates to verify this in CI, but it should be good to go.)

[nikita-volkov]

Cool. Released as 1.6.1

[robx]

It seems the build is broken:

library/Hasql/Private/Encoders.hs:236:17: error:
    • Couldn't match type ‘LazyByteString.ByteString’
                     with ‘ByteString’
      NB: ‘ByteString’ is defined in ‘Data.ByteString.Internal’
          ‘LazyByteString.ByteString’
            is defined in ‘Data.ByteString.Lazy.Internal’
      Expected type: Value ByteString
        Actual type: Value LazyByteString.ByteString
    • In the expression:
        Value (Value.unsafePTIWithShow PTI.json (const A.json_bytes_lazy))
      In an equation for ‘jsonLazyBytes’:
          jsonLazyBytes
            = Value
                (Value.unsafePTIWithShow PTI.json (const A.json_bytes_lazy))
    |
236 | jsonLazyBytes = Value (Value.unsafePTIWithShow PTI.json (const A.json_bytes_lazy))
    |                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

(see https://github.com/PostgREST/postgrest/runs/8247544920?check_suite_focus=true)

[nikita-volkov]

Oops. Forgot to check the build :) Fixed in 1.6.1.1