Re: segfault on rollback

Alvaro Herrera <alvherre(at)commandprompt(dot)com> writes:
> Tom Lane wrote:
>> This reminds me that even though the V3-protocol code has been in there
>> since 7.4, we don't have nearly complete test coverage on it. It'd be a
>> good idea to have some basic tests included somewhere in the standard or
>> contrib regression tests. Any thoughts about an easy way to do it?

> Do we have anything on core which speaks the v3 protocol? Since libpq
> doesn't expose functions to do all the stuff the v3 protocol can do, it
> would be difficult.

libpq doesn't expose everything in v3, but it exposes quite a lot ---
and will expose more once the Describe patch gets in. I think the main
thing you still can't do is issue a Bind as a separate operation from an
Execute, and that's not something I feel is real important for testing
the backend. The main problem is to have something that actually calls
the V3-protocol functions, which psql doesn't.

I was thinking of suggesting that even making psql use PQexecParams
instead of PQexec would help for testing purposes. But from a raw
performance point of view that's probably a loss; also it would be the
final blow to using a modern psql to connect to a pre-7.4 server.

regards, tom lane