(2013/07/20 1:11), Tom Lane wrote:
> Andres Freund <andres@2ndquadrant.com> writes:
>> On 2013-07-20 00:49:11 +0900, Hiroshi Inoue wrote:
>>> Using SnapshotSelf instead of SnapshotNow for currtid_ () wouldn't
>>> matter.
>
>> I think it actually might. You could get into dicey situations if you
>> use currtid_ in a query performing updates or inserts because it would
>> see the to-be-inserted tuple...
>
> I'm pretty sure Hiroshi-san was only opining about whether it would
> matter for ODBC's usage. IIUC, ODBC is using this function to re-fetch
> rows that it inserted, updated, or at least selected-for-update in a
> previous command of the current transaction, so actually any snapshot
> would do fine.
>
> In any case, since I moved the goalposts by suggesting that SnapshotSelf
> is just as dangerous as SnapshotNow, what we need to know is whether
> it'd be all right to change this code to use a fresh MVCC snapshot;
> and if not, why not. It's pretty hard to see a reason why client-side
> code would want to make use of the results of a non-MVCC snapshot.
OK I agree to replace SnapshotNow for currtid_xx() by a MVCC-snapshot.
regards,
Hiroshi Inoue
