Unable to encode array of composites

[bgamari]

The encoders supplied in Hasql.Encoders are not expressive enough to represent an array of composite values. Specifically, arrayValue expects a Value, which isn't a semigroup.

[nikita-volkov]

There's intentionally no encoders for composite types, because Postgres doesn't provide enough support for them (on the OID) level. Decoders are not a problem, but Encoders are.

[bgamari]

Nikita Volkov <[email protected]> writes:

There's intentionally no encoders for composite types, because Postgres doesn't provide enough support for them (on the OID) level.

What precisely is missing? Yes, you need to know the OID of the composite you are trying to construct, but it nevertheless should be possible to represent them.

[nikita-volkov]

There is no predefined OID for a composite type

[tekul]

Is there some way of working around this? I have a table which has a column containing an array of text, integer pairs and I'm not sure how to update or insert to that with Hasql. Would I have to build the explicit SQL for the entire command, including the array content?

[nikita-volkov]

Would I have to build the explicit SQL for the entire command, including the array content?

That's an option you always have.

As for the alternative, I recommend exploring something in the spirit of a solution to multiple values, based on unzipping of arrays. (Please, post back if you find a solution).

BTW, some form of support for composite encoding is likely to come in the next major release.

[tekul]

Thanks a lot for your super fast response! With the help of that link I managed to get it to work, though I'm pushing my boundaries somewhat, and there may be a better way :). Since my composite type only has two fields it's not overly complex. As suggested I split it into two separate arrays of type text and int2 and then recombine them with a cast in the SQL.

The relevant part of my schema is:

CREATE TYPE dict_entry AS (word text, index smallint);

CREATE TABLE story
    ( story_id text PRIMARY KEY
    , title text NOT NULL
    , img_url text NOT NULL
    , level smallint NOT NULL CHECK (level >= 0 AND level < 10)
    , curriculum text NOT NULL CHECK (length(curriculum) > 0)
    , tags text[] NOT NULL
    , content text NOT NULL CHECK (length(content) > 0)
    , words dict_entry[] NOT NULL
    , created_at timestamptz NOT NULL DEFAULT CURRENT_TIMESTAMP
    );

and my corresponding Hasql code is

evText :: E.Params Text
evText = E.value E.text

insertStory :: Query Story ()
insertStory = Q.statement sql storyEncoder D.unit True
  where
    sql = "INSERT INTO story (story_id, title, img_url, level, curriculum, tags, content, words) \
                 \VALUES ($1, $2, $3, $4, $5, $6, $7, (array(select word::dict_entry from unnest ($8, $9) as word)))"

storyEncoder :: E.Params Story
storyEncoder =
    contramap (fromJust . storyId) evText <>
    contramap title evText <>
    contramap img evText <>
    contramap (fromIntegral . storyLevel) (E.value E.int4) <>
    contramap (head . tags) evText <>
    contramap (tail . tags) (array E.text) <>
    contramap content evText <>
    contramap (map word . words) (array E.text) <>
    contramap (map (fromIntegral . index) . words) (array E.int2)
  where
    array v = E.value (E.array (E.arrayDimension foldl' (E.arrayValue v)))
    storyId = id :: Story -> Maybe Text
    storyLevel = level :: Story -> Int

The Haskell data types can be found here.

[nikita-volkov]

Yes. Seems like a good solution. Thanks for posting back!