sql: support * in udf bodies

This change allows `*` usage in UDF bodies. In order to facilitate
schema changes after a UDF is created but should not affect the UDF
output, we rewrite UDF ASTs in place to reference tables and columns by
ID instead of by name.

Informs: #90080

Epic: CRDB-19496
Release note (sql change): Allow `*` expressions in UDFs.

Author: rharding6373

pkg/ccl/backupccl/backup_test.go

@@ -10528,7 +10528,7 @@ $$;
 		require.Equal(t, 111, int(fnDesc.GetID()))
 		require.Equal(t, 104, int(fnDesc.GetParentID()))
 		require.Equal(t, 106, int(fnDesc.GetParentSchemaID()))
-		require.Equal(t, "SELECT a FROM db1.sc1.tbl1;\nSELECT nextval(110:::REGCLASS);", fnDesc.GetFunctionBody())
+		require.Equal(t, "SELECT a FROM [107(1) AS tbl1];\nSELECT nextval(110:::REGCLASS);", fnDesc.GetFunctionBody())
 		require.Equal(t, 100108, int(fnDesc.GetParams()[0].Type.Oid()))
 		require.Equal(t, []descpb.ID{107, 110}, fnDesc.GetDependsOn())
 		require.Equal(t, []descpb.ID{108, 109}, fnDesc.GetDependsOnTypes())
@@ -10581,7 +10581,7 @@ $$;
 		require.Equal(t, 112, int(fnDesc.GetParentID()))
 		require.Equal(t, 114, int(fnDesc.GetParentSchemaID()))
 		// Make sure db name and IDs are rewritten in function body.
-		require.Equal(t, "SELECT a FROM db1_new.sc1.tbl1;\nSELECT nextval(118:::REGCLASS);", fnDesc.GetFunctionBody())
+		require.Equal(t, "SELECT a FROM [107(1) AS tbl1];\nSELECT nextval(118:::REGCLASS);", fnDesc.GetFunctionBody())
 		require.Equal(t, 100116, int(fnDesc.GetParams()[0].Type.Oid()))
 		require.Equal(t, []descpb.ID{115, 118}, fnDesc.GetDependsOn())
 		require.Equal(t, []descpb.ID{116, 117}, fnDesc.GetDependsOnTypes())
@@ -10667,7 +10667,7 @@ $$;
 		require.Equal(t, 111, int(fnDesc.GetID()))
 		require.Equal(t, 104, int(fnDesc.GetParentID()))
 		require.Equal(t, 106, int(fnDesc.GetParentSchemaID()))
-		require.Equal(t, "SELECT a FROM db1.sc1.tbl1;\nSELECT nextval(110:::REGCLASS);", fnDesc.GetFunctionBody())
+		require.Equal(t, "SELECT a FROM [107(1) AS tbl1];\nSELECT nextval(110:::REGCLASS);", fnDesc.GetFunctionBody())
 		require.Equal(t, 100108, int(fnDesc.GetParams()[0].Type.Oid()))
 		require.Equal(t, []descpb.ID{107, 110}, fnDesc.GetDependsOn())
 		require.Equal(t, []descpb.ID{108, 109}, fnDesc.GetDependsOnTypes())
@@ -10722,7 +10722,7 @@ $$;
 		require.Equal(t, 107, int(fnDesc.GetParentID()))
 		require.Equal(t, 125, int(fnDesc.GetParentSchemaID()))
 		// Make sure db name and IDs are rewritten in function body.
-		require.Equal(t, "SELECT a FROM db1.sc1.tbl1;\nSELECT nextval(129:::REGCLASS);", fnDesc.GetFunctionBody())
+		require.Equal(t, "SELECT a FROM [107(1) AS tbl1];\nSELECT nextval(129:::REGCLASS);", fnDesc.GetFunctionBody())
 		require.Equal(t, 100127, int(fnDesc.GetParams()[0].Type.Oid()))
 		require.Equal(t, []descpb.ID{126, 129}, fnDesc.GetDependsOn())
 		require.Equal(t, []descpb.ID{127, 128}, fnDesc.GetDependsOnTypes())

pkg/ccl/backupccl/testdata/backup-restore/user-defined-functions

@@ -59,7 +59,7 @@ CREATE FUNCTION sc1.f1(IN a db1.sc1.enum1)
 	CALLED ON NULL INPUT
 	LANGUAGE SQL
 	AS $$
-	SELECT a FROM db1.sc1.tbl1;
+	SELECT a FROM [110(1) AS tbl1];
 	SELECT nextval('sc1.sq1'::REGCLASS);
 $$
 

pkg/ccl/logictestccl/tests/3node-tenant/generated_test.go

@@ -70,9 +70,9 @@ CREATE SCHEMA test_sc;
 		require.Equal(t, funcDesc.GetName(), "f")
 
 		require.Equal(t,
-			`SELECT a FROM defaultdb.public.t;
-SELECT b FROM defaultdb.public.t@t_idx_b;
-SELECT c FROM defaultdb.public.t@t_idx_c;
+			`SELECT a FROM [104(1, 2, 3) AS t];
+SELECT b FROM [104(1, 2, 3) AS t]@t_idx_b;
+SELECT c FROM [104(1, 2, 3) AS t]@t_idx_c;
 SELECT a FROM defaultdb.public.v;
 SELECT nextval(105:::REGCLASS);`,
 			funcDesc.GetFunctionBody())
@@ -255,7 +255,7 @@ $$;
 		require.Equal(t, funcDesc.GetName(), "f")
 
 		require.Equal(t,
-			`SELECT b FROM defaultdb.public.t1@t1_idx_b;
+			`SELECT b FROM [104(1, 2) AS t1]@t1_idx_b;
 SELECT a FROM defaultdb.public.v1;
 SELECT nextval(106:::REGCLASS);`,
 			funcDesc.GetFunctionBody())
@@ -298,7 +298,7 @@ $$;
 		require.Equal(t, funcDesc.GetName(), "f")
 
 		require.Equal(t,
-			`SELECT b FROM defaultdb.public.t2@t2_idx_b;
+			`SELECT b FROM [105(1, 2) AS t2]@t2_idx_b;
 SELECT a FROM defaultdb.public.v2;
 SELECT nextval(107:::REGCLASS);`,
 			funcDesc.GetFunctionBody())

pkg/sql/create_function_test.go

@@ -67,9 +67,9 @@ CREATE SCHEMA test_sc;
 		require.Equal(t, funcDesc.GetName(), "f")
 
 		require.Equal(t,
-			`SELECT a FROM defaultdb.public.t;
-SELECT b FROM defaultdb.public.t@t_idx_b;
-SELECT c FROM defaultdb.public.t@t_idx_c;
+			`SELECT a FROM [104(1, 2, 3) AS t];
+SELECT b FROM [104(1, 2, 3) AS t]@t_idx_b;
+SELECT c FROM [104(1, 2, 3) AS t]@t_idx_c;
 SELECT a FROM defaultdb.public.v;
 SELECT nextval(105:::REGCLASS);`,
 			funcDesc.GetFunctionBody())

pkg/sql/drop_function_test.go

@@ -120,9 +120,9 @@ CREATE FUNCTION f(INT) RETURNS INT IMMUTABLE LANGUAGE SQL AS $$ SELECT a FROM t
 
 		_, overload, err := funcResolver.ResolveFunctionByOID(ctx, funcDef.Overloads[0].Oid)
 		require.NoError(t, err)
-		require.Equal(t, `SELECT a FROM defaultdb.public.t;
-SELECT b FROM defaultdb.public.t@t_idx_b;
-SELECT c FROM defaultdb.public.t@t_idx_c;
+		require.Equal(t, `SELECT a FROM [104(1, 2, 3) AS t];
+SELECT b FROM [104(1, 2, 3) AS t]@t_idx_b;
+SELECT c FROM [104(1, 2, 3) AS t]@t_idx_c;
 SELECT a FROM defaultdb.public.v;
 SELECT nextval(105:::REGCLASS);`, overload.Body)
 		require.True(t, overload.IsUDF)
@@ -142,7 +142,7 @@ SELECT nextval(105:::REGCLASS);`, overload.Body)
 
 		_, overload, err = funcResolver.ResolveFunctionByOID(ctx, funcDef.Overloads[2].Oid)
 		require.NoError(t, err)
-		require.Equal(t, `SELECT a FROM defaultdb.public.t;`, overload.Body)
+		require.Equal(t, `SELECT a FROM [104(1, 2, 3) AS t];`, overload.Body)
 		require.True(t, overload.IsUDF)
 		require.False(t, overload.UDFContainsOnlySignature)
 		require.Equal(t, 1, len(overload.Types.Types()))

pkg/sql/function_resolver_test.go

@@ -52,24 +52,6 @@ CREATE FUNCTION err(i INT) RETURNS INT LANGUAGE SQL AS 'SELECT j'
 statement error pgcode 42703 column \"j\" does not exist
 CREATE FUNCTION err(i INT) RETURNS INT LANGUAGE SQL AS 'SELECT a FROM ab WHERE a = j'
 
-statement error pgcode 0A000 functions do not currently support \* expressions
-CREATE FUNCTION err(i INT) RETURNS ab LANGUAGE SQL AS 'SELECT * FROM ab'
-
-statement error pgcode 0A000 functions do not currently support \* expressions
-CREATE FUNCTION err(i INT) RETURNS ab LANGUAGE SQL AS 'SELECT ab.* FROM ab'
-
-statement error pgcode 0A000 functions do not currently support \* expressions
-CREATE FUNCTION err(i INT) RETURNS ab LANGUAGE SQL AS $$
-  SELECT 1;
-  SELECT * FROM ab;
-$$
-
-statement error pgcode 0A000 functions do not currently support \* expressions
-CREATE FUNCTION err(i INT) RETURNS INT LANGUAGE SQL AS $$
-  SELECT * FROM ab;
-  SELECT 1;
-$$
-
 statement ok
 CREATE FUNCTION d(i INT2) RETURNS INT4 LANGUAGE SQL AS 'SELECT i'
 
@@ -142,9 +124,9 @@ CREATE FUNCTION public.f(IN a test.public.notmyworkday)
   CALLED ON NULL INPUT
   LANGUAGE SQL
   AS $$
-  SELECT a FROM test.public.t;
-  SELECT b FROM test.public.t@t_idx_b;
-  SELECT c FROM test.public.t@t_idx_c;
+  SELECT a FROM [113(1, 2, 3) AS t];
+  SELECT b FROM [113(1, 2, 3) AS t]@t_idx_b;
+  SELECT c FROM [113(1, 2, 3) AS t]@t_idx_c;
   SELECT nextval('public.sq1'::REGCLASS);
 $$
 
@@ -2321,7 +2303,7 @@ CREATE FUNCTION public.get_l(IN i INT8)
     CALLED ON NULL INPUT
     LANGUAGE SQL
     AS $$
-    SELECT v FROM test.public.kv WHERE k = i;
+    SELECT v FROM [225(1, 2) AS kv] WHERE k = i;
 $$
 
 query T
@@ -2788,7 +2770,7 @@ SELECT oid, proname, pronamespace, proowner, prolang, proleakproof, proisstrict,
 FROM pg_catalog.pg_proc WHERE proname IN ('f_93314', 'f_93314_alias', 'f_93314_comp', 'f_93314_comp_t')
 ORDER BY oid;
 ----
-100257  f_93314         105  1546506610  14  false  false  false  v  0  100256  ·  {}  NULL  SELECT i, e FROM test.public.t_93314 ORDER BY i LIMIT 1;
-100259  f_93314_alias   105  1546506610  14  false  false  false  v  0  100258  ·  {}  NULL  SELECT i, e FROM test.public.t_93314_alias ORDER BY i LIMIT 1;
+100257  f_93314         105  1546506610  14  false  false  false  v  0  100256  ·  {}  NULL  SELECT i, e FROM [256(1, 2) AS t_93314] ORDER BY i LIMIT 1;
+100259  f_93314_alias   105  1546506610  14  false  false  false  v  0  100258  ·  {}  NULL  SELECT i, e FROM [258(1, 2) AS t_93314_alias] ORDER BY i LIMIT 1;
 100263  f_93314_comp    105  1546506610  14  false  false  false  v  0  100260  ·  {}  NULL  SELECT (1, 2);
-100264  f_93314_comp_t  105  1546506610  14  false  false  false  v  0  100262  ·  {}  NULL  SELECT a, c FROM test.public.t_93314_comp LIMIT 1;
+100264  f_93314_comp_t  105  1546506610  14  false  false  false  v  0  100262  ·  {}  NULL  SELECT a, c FROM [262(1, 2) AS t_93314_comp] LIMIT 1;

pkg/sql/logictest/testdata/logic_test/udf

@@ -0,0 +1,123 @@
+statement ok
+CREATE TABLE t_onecol (a INT);
+INSERT INTO t_onecol VALUES (1)
+
+statement ok
+CREATE TABLE t_twocol (a INT, b INT);
+INSERT INTO t_twocol VALUES (1,2)
+
+statement ok
+CREATE FUNCTION f_unqualified_onecol() RETURNS INT AS
+$$
+  SELECT * FROM t_onecol;
+$$ LANGUAGE SQL;
+
+statement ok
+CREATE FUNCTION f_subquery() RETURNS INT AS
+$$
+  SELECT * FROM (SELECT a FROM (SELECT * FROM t_onecol) AS foo) AS bar;
+$$ LANGUAGE SQL;
+
+statement ok
+CREATE FUNCTION f_unqualified_twocol() RETURNS t_twocol AS
+$$
+  SELECT * FROM t_twocol;
+$$ LANGUAGE SQL;
+
+statement ok
+CREATE FUNCTION f_allcolsel() RETURNS t_twocol AS
+$$
+  SELECT t_twocol.* FROM t_twocol;
+$$ LANGUAGE SQL;
+
+statement ok
+CREATE FUNCTION f_allcolsel_alias() RETURNS t_twocol AS
+$$
+  SELECT t1.* FROM t_twocol AS t1, t_twocol AS t2 WHERE t1.a = t2.a;
+$$ LANGUAGE SQL;
+
+statement ok
+CREATE FUNCTION f_tuplestar() RETURNS t_twocol AS
+$$
+  SELECT (t_twocol.*).* FROM t_twocol;
+$$ LANGUAGE SQL;
+
+query TTT
+SELECT oid, proname, prosrc
+FROM pg_catalog.pg_proc WHERE proname LIKE 'f\_%' ORDER BY oid;
+----
+100108  f_unqualified_onecol   SELECT * FROM [106(1) AS t_onecol];
+100109  f_subquery  SELECT * FROM (SELECT a FROM (SELECT * FROM [106(1) AS t_onecol]) AS foo) AS bar;
+100110  f_unqualified_twocol   SELECT * FROM [107(1, 2) AS t_twocol];
+100111  f_allcolsel   SELECT t_twocol.* FROM [107(1, 2) AS t_twocol];
+100112  f_allcolsel_alias   SELECT t1.* FROM [107(1, 2) AS t_twocol] AS t1, [107(1, 2) AS t_twocol] AS t2 WHERE t1.a = t2.a;
+100113  f_tuplestar   SELECT (t_twocol.*).* FROM [107(1, 2) AS t_twocol];
+
+query I
+SELECT f_unqualified_onecol()
+----
+1
+
+query I
+SELECT f_subquery()
+----
+1
+
+statement ok
+ALTER TABLE t_onecol ADD COLUMN b INT DEFAULT 5;
+
+query I
+SELECT f_unqualified_onecol()
+----
+1
+
+query I
+SELECT f_subquery()
+----
+1
+
+# It's ok to drop a column that was not used by the original UDF.
+statement ok
+ALTER TABLE t_onecol DROP COLUMN b;
+
+query T
+SELECT f_unqualified_twocol()
+----
+(1,2)
+
+query T
+SELECT f_allcolsel()
+----
+(1,2)
+
+query T
+SELECT f_allcolsel_alias()
+----
+(1,2)
+
+statement ok
+ALTER TABLE t_twocol ADD COLUMN c INT DEFAULT 5;
+
+# TODO(#95558): With early binding, postgres returns an error after adding a
+# column when the table is used as the return type. Note that this behavior is
+# ok for late binding.
+query T
+SELECT f_unqualified_twocol()
+----
+(1,2)
+
+# Altering a column type is not allowed in postgres or CRDB.
+statement error pq: cannot alter type of column "b" because function "f_unqualified_twocol" depends on it
+ALTER TABLE t_twocol ALTER b TYPE FLOAT;
+
+# TODO(harding): Postgres allows column renaming when only referenced by UDFs.
+statement error pq: cannot rename column "a" because function "f_unqualified_twocol" depends on it
+ALTER TABLE t_twocol RENAME COLUMN a TO d;
+
+# TODO(harding): Postgres allows table renaming when only referenced by UDFs.
+statement error pq: cannot rename relation "t_twocol" because function "f_unqualified_twocol" depends on it
+ALTER TABLE t_twocol RENAME TO t_twocol_prime;
+
+# Dropping a column a UDF depends on is not allowed.
+statement error pq: cannot drop column "b" because function "f_unqualified_twocol" depends on it
+ALTER TABLE t_twocol DROP COLUMN b;