It is my understanding that the goal of rel8 is to produce valid SQL in a type-safe manner. Meaning, if the Haskell code compiles without errors and does not use any constructs explicitly marked as "unsafe" in the library, then the produces SQL should always be valid (assuming the declared tables in code match the actual database schema).
I think I found a case where the produces SQL produces an error in Postgres before it is even checked against the schema.
I wrote this script which reproduces the issue:
#!/usr/bin/env stack
{- stack script
--resolver lts-20.3
--package rel8
-}
{-# LANGUAGE DeriveAnyClass #-}
{-# LANGUAGE DerivingStrategies #-}
import Data.Int (Int64)
import GHC.Generics (Generic)
import qualified Rel8
main = putStrLn $ Rel8.showQuery q
q =
Rel8.many (pure (Rel8.nothingTable :: Rel8.MaybeTable Rel8.Expr (Foo Rel8.Expr)))
>>= Rel8.catListTable
newtype Foo f = Foo {x :: Rel8.Column f Int64}
deriving stock Generic
deriving anyclass Rel8.Rel8ableThe script produces the following schema-independent SQL query:
SELECT
CAST("unnest0_4" AS bool) as "isJust",
CAST(CASE WHEN ("unnest0_4") IS NOT NULL THEN "unnest1_4" ELSE NULL END AS int8) as "Just/x"
FROM (SELECT
UNNEST(CASE WHEN ("rebind0_2") IS NULL THEN CAST(ARRAY[] AS bool[]) ELSE "result0_1" END) as "unnest0_4",
CASE WHEN (UNNEST(CASE WHEN ("rebind0_2") IS NULL THEN CAST(ARRAY[] AS bool[]) ELSE "result0_1" END)) IS NOT NULL THEN UNNEST(CASE WHEN ("rebind0_2") IS NULL THEN CAST(ARRAY[] AS int8[]) ELSE "result1_1" END) ELSE NULL END as "unnest1_4",
*
FROM (SELECT *
FROM
(SELECT
0) as "T1"
LEFT OUTER JOIN
(SELECT
TRUE as "rebind0_2",
*
FROM (SELECT
ARRAY_AGG("inner0_1") as "result0_1",
ARRAY_AGG("inner1_1") as "result1_1"
FROM (SELECT
CAST(NULL AS bool) as "inner0_1",
CASE WHEN (CAST(NULL AS bool)) IS NOT NULL THEN CAST(NULL AS int8) ELSE NULL END as "inner1_1",
*
FROM (SELECT
0) as "T1") as "T1"
GROUP BY COALESCE(0)) as "T1") as "T2"
ON
TRUE) as "T1") as "T1"When I try to execute the query I get the following error:
Query 1 ERROR: ERROR: argument of CASE/WHEN must not return a set
LINE 6: CASE WHEN (UNNEST(CASE WHEN ("rebind0_2") IS NULL THEN...
I think this indicates that there is a bug somewhere in rel8.
It is my understanding that the goal of rel8 is to produce valid SQL in a type-safe manner. Meaning, if the Haskell code compiles without errors and does not use any constructs explicitly marked as "unsafe" in the library, then the produces SQL should always be valid (assuming the declared tables in code match the actual database schema).
I think I found a case where the produces SQL produces an error in Postgres before it is even checked against the schema.
I wrote this script which reproduces the issue:
The script produces the following schema-independent SQL query:
When I try to execute the query I get the following error:
I think this indicates that there is a bug somewhere in rel8.