Re: [HACKERS] Errors inside transactions

"Thomas G. Lockhart" <lockhart(at)alumni(dot)caltech(dot)edu> writes:
> Why not have your error handler issue a new BEGIN statement? Or an
> END/BEGIN pair if that is what is needed? afaik that wouldn't be any
> different than the proposed "CONTINUE TRANSACTION" statement.
> Or do you really want to commit the previous statements in the
> transaction rather than throw them away? If so, you didn't really need
> the transaction block there anyway, since you are willing to accept a
> partial result.

You're right, I don't especially need or want the rollback aspect of
transactions. This is a shared database with half a dozen processes
watching it, and what I do want is to ensure that the other processes
will not see the logically-inconsistent state that occurs while updating
several related tuples with multiple UPDATE commands.

I guess what I'd really like is to be able to separate the atomic-update
and rollback features of transactions. If I could use "LOCK relation"
without having to be inside a transaction, I'd be a much happier camper.
(This'd at least imply an UNLOCK statement, I suppose. Are there deeper
relationships between locks and transactions?)

regards, tom lane