Adam Kennedy created CALCITE-6532:
-------------------------------------
Summary: Postgres RANDOM() is not flagged as non-deterministic
Key: CALCITE-6532
URL: https://issues.apache.org/jira/browse/CALCITE-6532
Project: Calcite
Issue Type: Bug
Reporter: Adam Kennedy
Putting aside the question of whether the default Calcite RAND can be
deterministic (when used with a seed) etc, it is most certainly the case that
the Postgres RANDOM() function is not deterministic.
The following should return two different values.
SELECT RANDOM(), RANDOM()
In Calcite, currently it will return the same value twice as RexProgramBuilder
deduplicates them to a single invocation (incorrectly).
This bug does not address the behavior of RexProgram, as there are separate
issues for the problem of RexProgram not obeying isDeterministic = false. But
were that bug to be fixed, it would be irrelevant to the SELECT RANDOM(),
RANDOM() problem as it stands, because a RexProgramBuilder that obey'ed
determinism wouldn't see RANDOM() as deterministic anyway.
Since RANDOM() is very new, and doesn't impact default Calcite use cases, I
propose that we can also make non-deterministic RANDOM() a trial run for a
later change of RAND() to non-deterministic.
It's not entirely clear to me what the semantic intent of isDynamic originally
was, I had assumed it was related to functions that need to reach outside the
context of the query to read from the environment, and that as a side effect
they should not be cached.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
