Andrew Dunstan <andrew(at)dunslane(dot)net> writes:
> Here's a patch I propose to apply to fix this bug (See
> <https://www.postgresql.org/message-id/flat/759e997e-e1ca-91cd-84db-f4ae963fada1%40dunslane.net#b1cf11c3eb1f450bed97c79ad473909f>)
If I'm reading the code correctly, your change in RelationBuildTupleDesc
is scribbling directly on the disk buffer, which is surely not okay.
I don't understand why you need that at all given the other defenses
you added ... but if you need it, you have to modify the tuple AFTER
copying it.
regards, tom lane