sql: use an implicit txn during the extended protocol

[rafiss]

fixes #71665

In the Postgres wire protocol, entering the extended protocol also
starts an implicit transaction. This transaction is committed when the
Sync message is received.

Previously, we were starting 3 separate "implicit" transactions for the
Prepare, Bind, and ExecPortal commands. (Implicit is in scare quotes
because they were created entirely ad-hoc.)

Now, the same implicit txn is used for the duration of the extended
protocol.

If BEGIN is executed as a prepared statement, then the implicit
trasnaction is upgraded to an explicit one. This behavior matches
Postgres.

Release note (bug fix): Fixed a bug where the different stages of
preparing, binding, and executing a prepared statement would use
different implicit transactions. Now these stages all share the same
implicit transaction.

[cockroach-teamcity]

This change is 

[rafiss]

Here's a test that fails before this PR, which I would have expected this PR to fix. But it still fails. The behavior of the last few lines is quite confusing -- preparing and executing the same statement with almost the same AOST timestamps gets different results. Maybe something to do with leases?

Anyway, unless someone feels strongly or has an idea for how to fix, I'll file that as a separate issue and leave it outside of this PR.

func TestResolveTypesAOST(t *testing.T) {
	srv, db, _ := serverutils.StartServer(t, base.TestServerArgs{Insecure: true})
	defer srv.Stopper().Stop(context.Background())
	defer db.Close()

	time.Sleep(2 * time.Second)

	_, err := db.Exec("CREATE TYPE typ AS ENUM('hi', 'hello')")
	require.NoError(t, err)

	var s string

	err = db.QueryRow("SELECT 'hi'::typ FROM generate_series(1,1) AS OF SYSTEM TIME '-1s'").Scan(&s)
	// Fails as expected
	require.EqualError(t, err, "pq: type \"typ\" does not exist")

	err = db.QueryRow("SELECT $1::typ FROM generate_series(1,1) AS OF SYSTEM TIME '-1s'", "hi").Scan(&s)
	// No error happens even though it should
	require.NoError(t, err)

	err = db.QueryRow("SELECT $1::typ FROM generate_series(1,1) AS OF SYSTEM TIME '-1s'", "hi").Scan(&s)
	// The error happens here though.
	require.EqualError(t, err, "pq: type \"typ\" does not exist")

	_, err = db.Exec("CREATE TABLE tab(a TEXT)")
	require.NoError(t, err)

	err = db.QueryRow("SELECT * FROM tab AS OF SYSTEM TIME '-1s'").Scan(&s)
	// No error happens even though it should
	require.NoError(t, err)

	err = db.QueryRow("SELECT * FROM tab AS OF SYSTEM TIME '-1s'").Scan(&s)
	// The error happens here though.
	require.EqualError(t, err, "pq: type \"typ\" does not exist")
}

[otan]

	err = db.QueryRow("SELECT * FROM tab AS OF SYSTEM TIME '-1s'").Scan(&s)
	// No error happens even though it should
	require.NoError(t, err)

	err = db.QueryRow("SELECT * FROM tab AS OF SYSTEM TIME '-1s'").Scan(&s)
	// The error happens here though.
	require.EqualError(t, err, "pq: type \"typ\" does not exist") 

why isn't the error "tab does not exist?"

[otan]

took a look, changes makes sense to me. will caveat i'm not 100% on the extended protocol. i trust ORM tests to break if something is off ;)

the error above is suspicious to me though.

[otan]

can you add a comment that the prepared statement is re-executed and we assume a constraint error from that? that confused me for a second.

[rafiss]

yeah for sure

[rafiss]

why isn't the error "tab does not exist?"

oops my bad -- copy paste error. also, it seems that one of the other fixes I made to clean up the PR made this test (with typos fixed) pass now, so we're all good!

[rafiss]

i trust ORM tests to break if something is off ;)

I'll try running the ORM tests before merging this

[rafiss]

tftr!

bors r=otan

[craig]

Build succeeded:

• GitHub CI (Cockroach)