sql: populate information_schema.user_defined_types and attributes

[annrpom]

commit 1-4:

sql: add necessary builtins for information_schema.attributes

These patches add the following builtins:

• information_schema._pg_char_octet_length: returns the octet length of a given char/varchar/string
• pg_encoding_max_length: returns the maximum length of a given encoding. Since we only support UTF8, this will return length 4.
• information_schema._pg_datetime_precision: returns the precision of the given datetime-related type
• information_schema._pg_interval_type: displays the interval type duration if it is set

Needed for: #109603
Epic: none

Release note (sql change): The information_schema._pg_char_octet_length, information_schema._pg_datetime_precision, information_schema._pg_interval_type, and pg_encoding_max_length builtin functions are now supported, which improves compatibility with PostgreSQL.

---

last commit:

sql: populate information_schema.user_defined_types and attributes

This patch populates information_schema.user_defined_types with
information about user defined types and information_schema.attributes
with information about the attributes of composite data types.

Epic: none
Fixes: #109603

Release note (sql change): This patch populates
information_schema.user_defined_types with information about user
defined types and information_schema.attributes with information about
the attributes of composite data types.

[cockroach-teamcity]

This change is 

[annrpom]

• add tests udt and attribute tables
• add tests for builtins

ideally, i get #111179 merged so that i can write tests and check them locally properly

[rafiss]

this is great! my comments are all minor

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

---

pkg/sql/sem/builtins/pg_builtins.go line 2186 at r1 (raw file):

	),

	// NOTE: this could be defined as a user-defined function, like

nit: this comment seems outdated, since that's how the function is defined now. we could just keep the part that links to postgres source

---

pkg/sql/sem/builtins/pg_builtins.go line 2203 at r1 (raw file):

	//          ELSE null
	// END;
	"information_schema._pg_char_octet_length": makeBuiltin(tree.FunctionProperties{Category: builtinconstants.CategorySystemInfo},

this function could use some basic tests

---

pkg/sql/sem/builtins/pg_builtins.go line 2212 at r1 (raw file):

			Body: `SELECT
						 CASE WHEN $1 IN (25, 1042, 1043)
			            THEN CASE WHEN $2 = -1

nit: the comments that are included above would make this more understandable. i mean the /* text, char, varchar */ and /* default typmod */ comments

---

pkg/sql/sem/builtins/pg_builtins.go line 2273 at r3 (raw file):

	),

	// NOTE: this could be defined as a user-defined function, like

nit: "could be" is outdated here; this comment just needs a link to the postgres code

---

pkg/sql/sem/builtins/pg_builtins.go line 2294 at r3 (raw file):

			ReturnType: tree.FixedReturnType(types.String),
			Body: `SELECT
						 CASE WHEN $1 IN (1186) 

nit: the /* interval */ comment could be helpful here

---

pkg/sql/logictest/testdata/logic_test/pg_builtins line 951 at r3 (raw file):

ORDER BY a.attname
----
attname  typname   information_schema._pg_interval_type

nit: could you add some test cases where the result of `pg_interval_type is not null?

[annrpom]

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

---

pkg/sql/sem/builtins/pg_builtins.go line 2186 at r1 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

nit: this comment seems outdated, since that's how the function is defined now. we could just keep the part that links to postgres source

done

---

pkg/sql/sem/builtins/pg_builtins.go line 2203 at r1 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

this function could use some basic tests

these were added in the commit with pg_encoding_max_length - since pg_encoding_max_length was needed for _pg_char_octet_length

i added some tests for pg_encoding_max_length and moved the commits next to eachother!

---

pkg/sql/sem/builtins/pg_builtins.go line 2212 at r1 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

nit: the comments that are included above would make this more understandable. i mean the /* text, char, varchar */ and /* default typmod */ comments

done

---

pkg/sql/sem/builtins/pg_builtins.go line 2273 at r3 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

nit: "could be" is outdated here; this comment just needs a link to the postgres code

done

---

pkg/sql/sem/builtins/pg_builtins.go line 2294 at r3 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

nit: the /* interval */ comment could be helpful here

done

---

pkg/sql/logictest/testdata/logic_test/pg_builtins line 951 at r3 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

nit: could you add some test cases where the result of `pg_interval_type is not null?

hmmm looks like this will return null for now due to how interval types in TypeModifier get handled - let me get a fix up for this. think I need to have another check for if t.InternalType.IntervalDurationField.DurationType != IntervalDurationType_UNSET - not sure yet what I should be returning. maybe postgres source code has something for me

[annrpom]

maybe postgres source code has something for me

This is now looking like a separate issue - as this looks to be a compatibility bug that existed before this PR (see issue above) For now, information_schema._pg_interval_type does not work properly on intervals that have durations on them

[rafiss]

lgtm!! nice work on seeing this through. (feel free to rebase and merge)

Reviewable status: complete! 0 of 0 LGTMs obtained

[annrpom]

TFTR! ('-')7

bors r=rafiss

[annrpom]

bors r-

[craig]

Canceled.

[annrpom]

bors r+

[craig]

Build succeeded:

• Bazel Essential CI (Cockroach)