sql: update UDF references in ALTER TABLE

[chrisseto]

Previously, ALTER TABLE ... ADD CONSTRAINT statements that referenced a UDF did not appropriately update UDFs' DependedOnBy field to reference the table/constraint that utilized it. This would surface as the validation error reported in #109414, relation X: depends-on function X has no corresponding depended-on-by back reference. This commit ensures that the back references are updated by copying a hook from CREATE TABLE.

Epic: None
Fixes: #109414
Release note (bug fix): ALTER TABLE ... ADD CONSTRAINT CHECK ... statements that utilize a UDF in the CHECK no longer cause a validation error.

[cockroach-teamcity]

This change is 

[rafiss]

looks great! just had very minor testing comments

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @chrisseto)

---

pkg/sql/logictest/testdata/logic_test/udf_in_constraints line 444 at r1 (raw file):

subtest issue-109414-minimal

statement ok

very cool that we got a repro for this!

one thing which may not be obvious is that if you send all the queries at once like this, they are al run in the same transaction (it's a PG compat thing). to make this more clear, could we

• add a BEGIN/COMMIT around the whole thing, so it's clear it's all the same transaction
• or add a BEGIN/COMMIT around each command, so each one is forced to run in a different transaction

---

pkg/sql/logictest/testdata/logic_test/udf_in_constraints line 448 at r1 (raw file):

CREATE FUNCTION has_account(account_id UUID) RETURNS BOOL LANGUAGE SQL AS $$ SELECT EXISTS(SELECT * FROM accounts WHERE id = account_id) $$;
CREATE TABLE account_ref ( account_id UUID NOT NULL );
ALTER TABLE account_ref ADD CONSTRAINT diy_fk CHECK (has_account(account_id));

nit: it might be useful to throw in an assertion here, like check the output of SHOW CREATE TABLE account_ref

---

pkg/sql/logictest/testdata/logic_test/udf_in_constraints line 453 at r1 (raw file):

# of itself, so it's been included as it may catch other regressions that the
# minimal case wouldn't.
subtest issue-109414-full

both comments above apply to this subtest too

[chrisseto]

TFTR!

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @rafiss)

---

pkg/sql/logictest/testdata/logic_test/udf_in_constraints line 444 at r1 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

very cool that we got a repro for this!

one thing which may not be obvious is that if you send all the queries at once like this, they are al run in the same transaction (it's a PG compat thing). to make this more clear, could we

• add a BEGIN/COMMIT around the whole thing, so it's clear it's all the same transaction
• or add a BEGIN/COMMIT around each command, so each one is forced to run in a different transaction

done and I managed to simplify the reproduction a bit further.

---

pkg/sql/logictest/testdata/logic_test/udf_in_constraints line 448 at r1 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

nit: it might be useful to throw in an assertion here, like check the output of SHOW CREATE TABLE account_ref

Done.

---

pkg/sql/logictest/testdata/logic_test/udf_in_constraints line 453 at r1 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

both comments above apply to this subtest too

Done.

[rafiss]

nice work!

Reviewable status: complete! 0 of 0 LGTMs obtained

[rafiss]

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @chrisseto)

---

pkg/sql/logictest/testdata/logic_test/udf_in_constraints line 469 at r2 (raw file):

statement ok
CREATE TABLE accounts_a (id UUID NOT NULL);
CREATE TABLE accounts_b (id UUID NOT NULL);

ah, the test is failing due to something the logic tests do called column family randomization. basically, it will add non-default column families, so the output of SHOW is affected. you can fix it by making the CREATE statement be explicit about column families, like this:

CREATE TABLE accounts_a (id UUID NOT NULL, FAMILY "primary" (id, rowid))

[chrisseto]

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @rafiss)

---

pkg/sql/logictest/testdata/logic_test/udf_in_constraints line 469 at r2 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

ah, the test is failing due to something the logic tests do called column family randomization. basically, it will add non-default column families, so the output of SHOW is affected. you can fix it by making the CREATE statement be explicit about column families, like this:

CREATE TABLE accounts_a (id UUID NOT NULL, FAMILY "primary" (id, rowid))

aaaaah, thank you! Fixed.

[chrisseto]

bors r=rafiss

[craig]

Build succeeded:

• Bazel Essential CI (Cockroach)