I forget or do not notice some of your questions, would you be so kind as to repeat them?
Sent early by accident.
Repeating questions:
Why do you need to do all this indirection with changing RelFileNode to RelFileNodeBackend in the bufmgr, changing BufferGetTag etc? Similarly, your changes of RelFileNodeBackendIsTemp to RelFileNodeBackendIsLocalTemp . I'm guessing you did it the way you did instead to lay the groundwork for cross-backend sharing, but if so it should IMO be in your second patch that adds support for using shared_buffers for temp tables, not in the first patch that adds a minimal global temp tables implementation. Maybe my understanding of the existing temp table mechanics is just insufficient as I see RelFileNodeBackendIsTemp is already used in some aspects of existing temp relation handling.
Did you look into my suggestion of extending the relmapper so that global temp tables would have a relfilenode of 0 like pg_class etc, and use a backend-local map of oid-to-relfilenode mappings?
Similarly, TruncateSessionRelations probably shouldn't need to exist in this patch in its current form; there's no shared_buffers use to clean and the same file cleanup mechanism should handle both session-temp and local-temp relfilenodes.