On 9/21/07, Christopher Browne <cbbrowne at ca.afilias.info> wrote:
> Marko Kreen wrote:
> > On 9/21/07, Christopher Browne <cbbrowne at ca.afilias.info> wrote:
> >> The only thing that might (or mightn't!) be controversial is that it's
> >> pretty vital that the implementation remains consistent.
> >
> > What inconsistencies are you thinking of?
>
> Anything that might cause one implementation to differ from another.
> (e.g. - where Skytools and Slony-I might have differing code)
>
> I have no particular inconsistency in mind; I don't expect any.
>
> I don't think there's any *good* reason for them to differ; I'd hope
> they wouldn't differ.  They can't, otherwise the separate versions
> would need to be retained.

Yes, I also don't see any need for differences.

Only question that may need discussion is the question of naming.

Current APi is following:

 type: txid_snapshot

 get_current_txid(): int8
 get_current_snapshot(): txid_snapshot
 get_snapshot_xmin(sn): int8
 get_snapshot_xmax(sn): int8
 get_snapshot_active(sn): setof int8
 txid_in_snapshot(int8, sn): bool

But it seems kind of random, considering it is targeted to be core
functinality in the future.  I'd prefer something more concise:

 type: snapshot

 current_txid(): int8
 current_snapshot(): snapshot
 snapshot_xmin()
 snapshot_xmax()
 snapshot_uncommitted()  ??
 snapshot_contains()     ??

Somewhat better but the last 2 functions are still dubious.
Do you have better names?

Another thing - maybe they should be prefixed with pg_ considering
they are pretty lowlevel things?  But this call should probably
made by pg-hackers.  If we have some consensus here, then we
should probably ping them also.

It would be good to get the final API soon, so that whenever
the final merge happens, it would not mean API breakage.  Then
we can start coding against final API immidiately.

-- 
marko