Re: Rollback on Error

Tom Lane wrote:

> "Michael Paesold" <mpaesold(at)gmx(dot)at> writes:
> > On the other hand, the scenario of a psql option (read: I have
> > given up the idea of a backend implementation) to rollback only
> > last statement on error is quite different.
>
> Sure (and we already have one for autocommit). But I thought you were
> asking about a backend implementation.

I have implemented what I have suggested for psql. I have attached a first
patch for review here, because I have a few questions. Also I want to make
sure the whole thing is reasonable.

I have named the option "IMPLICIT_SAVEPOINTS", because that's what it is. If
someone has a better name that would describe the purpose of the feature, I
am happy to change it.

The feature is activated, if
* \set IMPLICIT_SAVEPOINTS 'on'
* connection is in "idle in transaction" state
* psql session is interactive

The code executes an implicit "SAVEPOINT pg_internal_psql" in
common.c/SendQuery to which it will try to rollback to, if the executed
query fails.

Open questions:
* Should psql print a notice in the case of that rollback?
Something like "Rollback of last statement successful."?

* What is currently missing, is a detection of \i ... obviously this feature
should not be used for each query in \i. Perhaps only for the whole \i
command?
So what should I do to detect \i?
Add an extra argument to MainLoop, SendQuery and process_file()? (many
changes)
Add a global variable in common.c/h (e.g. bool
deactivate_implicit_savepoints) that can be used in process_file to
temporarily deactivate the code path?
(more local changes, but rather a hack imho)

Please have a look at the patch and comment.

Best Regards,
Michael Paesold