-- +goose Up
-- +goose StatementBegin
-- A CPV code has one official description, so it is now one row whatever
-- description it arrives with: EU Supply cuts descriptions to 64 characters,
-- which made a second row for each code it sent. The duplicates are merged
-- into the row with the longest description, and the upsert keeps the
-- longest from now on. Other schemes still get a row per distinct content.

-- The row each CPV row merges into.
CREATE TEMP TABLE "cpv_merge" AS
SELECT
	"c"."id" AS "id",
	(
		SELECT "k"."id" FROM "classification_codes" "k"
		WHERE "k"."is_cpv" AND "k"."code" = "c"."code"
		ORDER BY length("k"."classification") DESC, "k"."id"
		LIMIT 1
	) AS "keep_id"
FROM "classification_codes" "c"
WHERE "c"."is_cpv";

-- An opportunity linked to both rows keeps only its link to the kept one.
DELETE FROM "opportunity_classifications"
WHERE "classification_id" IN (
	SELECT "id" FROM "cpv_merge" WHERE "id" <> "keep_id"
)
AND EXISTS (
	SELECT 1
	FROM "cpv_merge" "m"
	JOIN "opportunity_classifications" "kept"
		ON "kept"."classification_id" = "m"."keep_id"
	WHERE "m"."id" = "opportunity_classifications"."classification_id"
		AND "kept"."opportunity_uuid"
			= "opportunity_classifications"."opportunity_uuid"
);

UPDATE "opportunity_classifications"
SET "classification_id" = (
	SELECT "keep_id" FROM "cpv_merge"
	WHERE "id" = "opportunity_classifications"."classification_id"
)
WHERE "classification_id" IN (
	SELECT "id" FROM "cpv_merge" WHERE "id" <> "keep_id"
);

DELETE FROM "classification_codes"
WHERE "id" IN (SELECT "id" FROM "cpv_merge" WHERE "id" <> "keep_id");

DROP TABLE "cpv_merge";

CREATE UNIQUE INDEX IF NOT EXISTS "classification_codes_cpv"
	ON "classification_codes" ("code") WHERE "is_cpv";
-- +goose StatementEnd

-- +goose Down
-- +goose StatementBegin
-- The merged duplicates aren't restored: they only differed by a cut-off
-- description.
DROP INDEX IF EXISTS "classification_codes_cpv";
-- +goose StatementEnd