Closed WinterUnicorn closed 1 year ago
I am not completely sure of this consequences that this change has, but shouldn't we have at least a regression test to track this behavior? A backpatch could create a change of behavior for existing users, as well.
I tried the repro with master (PG15) branch but failed. Error-return from nested planner is checked in ut-A but actually error-case is not tested (ut-A). (opposed to my memory)
So, +1 for adding the test for that case. That part is repeatedly rewrote seeking more stable behavior but..
+1 for adding test cases. We also need to update the documentation.
Adding '[' and ']' to the white list seems reasonable to me since these characters can also be used as data types in PREPARE statement. But it would change the existing behavior so not sure we should back-patch it.
And so, I have added a regression test and applied that on HEAD as of 7c03155 with no backpatch involved so this will be available in the next versions compatible with Postgres 15~.
We have an issues with prepared statements that use arrays as arguments. In this case pg_hint_plan don't pass preceding sql check.
Example:
This request to add the '[', ']' symbols into the white list.