Re: pgsql: Fix memory leak in pgbench

>> I can set up farm animals with bleeding age clang/gcc (compiled weekly from
>> sources) with -Werror (seawasp & moonjelly are already running with those
>> but without -Werror).
>
> If possible, I would not recommend using compiled versions from
> source, but just the latest stable versions released for gcc and
> clang.

Yep, but I'm a compiler person, so I'm happy with compiling bleeding edge
pg with bleeding edge gcc & clang.

Note that there is also "caiman" running a non release gcc version.

> Even if this can catch up compiler bugs using Postgres code at
> a very early stage,

Indeed, several times, and they were reported and fixed.

> the false positives induced have been proving to hurt because of the
> time investigations took to determine if this was a problem on our side
> or not.

ISTM that the answer is that they are experimental animals, hence the
choice of stingy jellyfish names, and failures should not be investigated
immediately.

Maybe some "experimental" warning could be shown somewhere as a reminder?
I can add it in the compiler version, something like:

"Ubuntu 18.04.2 LTS **EXPERIMENTAL** gcc ..."

--
Fabien.