Attached test case demonstrates an issue with nbtree's mark/restore
code. All supported versions are affected.
My suspicion is that bugfix commit 70bc5833 missed some subtlety
around what we need to do to make sure that the array keys stay "in
sync" with the scan. I'll have time to debug the problem some more
tomorrow.
My ScalarArrayOp project [1] seems unaffected by the bug, so I don't
expect it'll take long to get to the bottom of this. This is probably
due to its general design, and not any specific detail. The patch
makes the relationship between the current scan position and the
current array keys a great deal looser.
[1] https://commitfest.postgresql.org/44/4455/
--
Peter Geoghegan