Revisited: Does error within transaction imply restarting it?

Hi everybody,

I'm the one seeking PostgreSQL not to fall in *ABORT STATE* after an erroneous
command within a transaction. Peter Eisentrout, suggested a couple of months
ago, to hack the backend in a very simple way (deleting the line which aborted
after an error). This worked well for me in 6.5.3 as far as I could test. But in
7.0, this doesn't perform well, as some other folk already reported. I'm willing
to do as much work as I can towards getting this behaviour change as soon as
possible. I would like you all to help me starting a discussion, where I could
find the central issues which will allow to get this done.

Specifically, the problem which generates deleting the above mentioned line, is
corrupted shared memory, and backend chrash and restart.

Any suggestions?

Regards,
Haroldo Stenger.