sql: implement RETURNS TABLE syntax

[DrewKimball]

sql: do not limit set-returning UDF when it has OUT parameters

This commit fixes a bug that caused the SetOf option for the UDF
ReturnType to be overwritten if the UDF had OUT parameters. The bug
caused a LIMIT 1 to be imposed on the UDF's final body statement, so
that the UDF returned only a single row.

Fixes #128403

Release note (bug fix): Fixed a bug existing since v24.1 that would
cause a set-returning UDF with OUT parameters to return a single row.

sql: implement RETURNS TABLE syntax

This commit implements RETURNS TABLE for UDFs. RETURNS TABLE is
syntactic sugar for RETURNS SETOF with:

• RECORD if there are multiple TABLE parameters, or
• the type of the single TABLE parameter.
  The TABLE parameters are added to the list of routine parameters.

Fixes #100226

Release note (sql change): Added support for RETURNS TABLE syntax when
creating a UDF.

[cockroach-teamcity]

This change is 

[rytaft]

Do you plan to backport just the first commit? Or both?

[DrewKimball]

Do you plan to backport just the first commit? Or both?

Just the first commit.

[yuzefovich]

Nice! Just some nits from me.

Reviewed 3 of 3 files at r1, 6 of 6 files at r2, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @DrewKimball and @mgartner)

---

pkg/sql/logictest/testdata/logic_test/udf_setof line 246 at r2 (raw file):

subtest returns_table

statement error pgcode 42601 pq: at or near "EOF": syntax error: OUT and INOUT arguments aren't allowed in TABLE functions

nit: it seems a bit dirty that we have at or near "EOF": syntax error: prefix in the error message - the statement is parsable. Should we do this check after parsing for a cleaner message?

---

pkg/sql/opt/optbuilder/create_function.go line 310 at r1 (raw file):

			// errors in shared logic later.
			funcReturnType = types.Void
			cf.ReturnType = &tree.RoutineReturnType{

nit: perhaps add an test-only assertion that for procedures SetOf is not set?

---

pkg/sql/parser/sql.y line 4892 at r2 (raw file):

    // - RECORD if there are multiple TABLE parameters, or
    // - the type of the single TABLE parameter.
    // The TABLE parameters are

nit: the comment seems incomplete.

[DrewKimball]

TFTR!

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @mgartner and @yuzefovich)

---

pkg/sql/opt/optbuilder/create_function.go line 310 at r1 (raw file):

Previously, yuzefovich (Yahor Yuzefovich) wrote…

nit: perhaps add an test-only assertion that for procedures SetOf is not set?

For procedures, ReturnType isn't set at all until this point.

---

pkg/sql/parser/sql.y line 4892 at r2 (raw file):

Previously, yuzefovich (Yahor Yuzefovich) wrote…

nit: the comment seems incomplete.

Done.

---

pkg/sql/logictest/testdata/logic_test/udf_setof line 246 at r2 (raw file):

Previously, yuzefovich (Yahor Yuzefovich) wrote…

nit: it seems a bit dirty that we have at or near "EOF": syntax error: prefix in the error message - the statement is parsable. Should we do this check after parsing for a cleaner message?

Done. It's definitely easiest to emit the error while parsing, because we parse functions in multiple places, and it means we don't have to track table function parameters separately in the AST. I added lexer.setErrNoDetails for cases like this where we don't want to annotate the error beyond giving it the Syntax error code.

[DrewKimball]

TFTR!

bors r+

[craig]

Build failed:

• macos_arm64_build

[DrewKimball]

bors retry

[craig]

Build succeeded:

• linux_arm64_build
• docker_image_amd64
• unit_tests
• check_generated_code
• macos_arm64_build
• local_roachtest
• linux_amd64_build
• examples_orms
• windows_build
• lint
• local_roachtest_fips
• acceptance
• macos_amd64_build
• linux_amd64_fips_build