I wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> I think the problem is that load_enum_cache_data() uses
>> GetTransactionSnapshot() rather than GetLatestSnapshot().
> That would only make the race condition window smaller (ie, hard
> to reproduce manually like this, but not gone).
No, wait, we made ALTER TYPE ADD VALUE PreventTransactionChain so that
uncommitted enum OIDs could never get into tables or indexes. So I
think you're right, forcing a new snapshot to be used would fix this.
However, I'm a bit worried by the "if (!FirstSnapshotSet)" restriction
in GetLatestSnapshot. Are we sure that enum comparisons could never
happen without a snapshot already being set? What's the point of
throwing an error there anyway, as opposed to letting it redirect to
GetTransactionSnapshot?
regards, tom lane