On 10/1/07, Tom Lane <tgl at sss.pgh.pa.us> wrote:
> "Marko Kreen" <markokr at gmail.com> writes:
> > On 10/1/07, Tom Lane <tgl at sss.pgh.pa.us> wrote:
> >> I'm having quite a bit of a problem with the above.  Why is
> >> InvalidTransactionId mapped to MAX_TXID, which presumably is part of the
> >> normal XID rotation and hence only larger than half of the universe,
> >> when the other special XIDs map as themselves?
>
> > Because InvalidTransactionId is supposed to be always invisible,
> > but others always visible?  And we don't want to add epoch to
> > any of them.
>
> Well, all three of the "special" xids need to be epoch-independent.
> I still think this is either bad design or an outright bug.

I don't understand you.  Current code _is_ making them
epoch-independent?  What aspect of it do you dislike?

> > I used StringInfo as it was only buffer tool available from backend...
>
> Maybe, but don't be too surprised if it breaks under you ...

What problems do you see?  I can do explicit buffer management
if it's unreliable for binary data handling.

-- 
marko