roachtest/sqlsmith: error decoding 10 bytes: empty array

[yuzefovich]

Extracted from #70666 (comment).

roachtest.sqlsmith/setup=empty/setting=no-ddl failed with artifacts on master @ ecaa4920192549af1418923648769ede3de1c9c8:

The test failed on branch=master, cloud=gce:
test artifacts and logs in: /home/agent/work/.go/src/github.com/cockroachdb/cockroach/artifacts/sqlsmith/setup=empty/setting=no-ddl/run_1
	sqlsmith.go:223,sqlsmith.go:256,test_runner.go:777: error: pq: internal error: unexpected error from the vectorized engine: error decoding 10 bytes: empty array
		stmt:
SELECT
  col
FROM
  (VALUES ((ARRAY[]:::RECORD[],)), ((ARRAY[()],)))
    AS t(col)
ORDER BY
  col ASC;

Reproduce

See: roachtest README

This test on roachdash | Improve this report!

[yuzefovich]

I've debugged this a bit, and the issue is that we have two datums with slightly different types but are assuming that they have exactly the same types during the decoding process.

In the image below, it is a bit hard to see, but

• the first datum is a tuple consisting of a single array element, with array's element being a tuple that itself has a single element (i.e. `len(TupleContents) == 1)
• the second datum is a tuple consisting of a single array element, with array's element being a tuple that itself has no elements (i.e. `len(TupleContents) == 0)
• during the decoding of the empty "inner" tuple in the second datum we get the decoding error.

I'm not sure whether we want to simply prohibit such cases during the type checking or not. FWIW the postgres returns a syntax error for the reproduction in this issue.

I'll assign this to @rafiss given that in #63996 he touched the type checking of arrays of tuples.

[rafiss]

Postgres doesn't allow the empty tuple () in this context, but here's a case that it does handle, but CRDB errors with a similar stack trace as originally reported.

SELECT
  col
FROM
  (VALUES ((ARRAY[]::RECORD[])), ((ARRAY[(1,2)])))
    AS t(col)
ORDER BY
  col ASC;

The problem is that the first array is an array of AnyTuples. and AnyTuple has a TupleContents equal to {types.Any}. The decoding logic gets confused, since it just thinks the tuple it's decoding should always have 1 element in it.