sqlsmith: add support for creating udfs

[rharding6373]

This PR gives sqlsmith the capability to issue CREATE FUNCTION commands. It randomly chooses from available and valid function options. It also contains up to 10 SELECT statements in the UDF body.

This PR also leaves several TODOs, including support for params, return types other than records, and calling UDFs in queries after they are created.

Epic: CRDB-20370
Informs: #90782

Release note: None

[cockroach-teamcity]

This change is 

[jayshrivastava]

Approving to unblock you from my previous review

[mgartner]

Nice!

Reviewed 3 of 3 files at r1, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @DrewKimball, @rharding6373, and @shermanCRL)

---

-- commits line 6 at r1:
nit: I believe it's up to 9 since IntN(10) returns numbers in the rang [0, 10).

---

pkg/internal/sqlsmith/relational.go line 910 at r1 (raw file):

	// 15%: FunctionCalledOnNullInput
	// 15%: FunctionReturnsNullOnNullInput
	// 15%: FunctionStrict

nit: Maybe this should be ~15% - I was a bit confused why they didn't add up to 100%.

---

pkg/internal/sqlsmith/relational.go line 924 at r1 (raw file):

	// 15%: FunctionVolatile
	// 15%: FunctionImmutable
	// 15%: FunctionStable

nit: ~15%

---

pkg/internal/sqlsmith/relational.go line 963 at r1 (raw file):

	for i := 0; i < stmtCnt; i++ {
		// UDFs currently only support SELECT statements.
		stmt, _, ok := s.makeSelect(nil, nil)

nit: label these nils

---

pkg/internal/sqlsmith/relational.go line 967 at r1 (raw file):

			continue
		}
		stmts = append(stmts, stmt.String())

nit: I think we might have to use tree.AsStringWithFlags with the tree.FmtParsable flag to ensure these statements can be parsed.

---

pkg/internal/sqlsmith/relational.go line 970 at r1 (raw file):

	}
	if len(stmts) == 0 {
		return nil, false

UDFs with empty bodies are allowed. They're not that interesting but I think they're worth testing, at least some of the time.

[DrewKimball]

Nice work!

Reviewed 3 of 3 files at r1, all commit messages.
Reviewable status: complete! 2 of 0 LGTMs obtained (waiting on @rharding6373 and @shermanCRL)

[rharding6373]

TFTRs!

Reviewable status: complete! 0 of 0 LGTMs obtained (and 2 stale) (waiting on @DrewKimball, @jayshrivastava, @mgartner, and @shermanCRL)

---

-- commits line 6 at r1:

Previously, mgartner (Marcus Gartner) wrote…

nit: I believe it's up to 9 since IntN(10) returns numbers in the rang [0, 10).

Good point. I increased the range to 11 to make this true.

---

pkg/internal/sqlsmith/relational.go line 910 at r1 (raw file):

Previously, mgartner (Marcus Gartner) wrote…

nit: Maybe this should be ~15% - I was a bit confused why they didn't add up to 100%.

Changed to ~17%. Closer to 100, but now going over.

---

pkg/internal/sqlsmith/relational.go line 924 at r1 (raw file):

Previously, mgartner (Marcus Gartner) wrote…

nit: ~15%

Done.

---

pkg/internal/sqlsmith/relational.go line 963 at r1 (raw file):

Previously, mgartner (Marcus Gartner) wrote…

nit: label these nils

Done.

---

pkg/internal/sqlsmith/relational.go line 967 at r1 (raw file):

Previously, mgartner (Marcus Gartner) wrote…

nit: I think we might have to use tree.AsStringWithFlags with the tree.FmtParsable flag to ensure these statements can be parsed.

Surprisingly didn't seem to be an issue when stressing TestGenerateParse, but this seems like a good idea.

---

pkg/internal/sqlsmith/relational.go line 970 at r1 (raw file):

Previously, mgartner (Marcus Gartner) wrote…

UDFs with empty bodies are allowed. They're not that interesting but I think they're worth testing, at least some of the time.

Done.

---

pkg/ccl/changefeedccl/changefeed_test.go line 0 at r1 (raw file):

Previously, jayshrivastava (Jayant) wrote…

Sounds good. This LGTM from the CDC side

Thank you for taking a look!

[rharding6373]

bors r+

[craig]

Build succeeded:

• Bazel Essential CI (Cockroach)