On Tue, 27 Feb 2024 at 00:31, Ivan Trofimov <i.trofimow@yandex.ru> wrote:
> I see now that I failed to express myself clearly: it's not a per-query
> overhead, but rather a per-result-field one.
I'm fairly sympathetic to decreasing the overhead of any per-row
operation. And looking at the code, it doesn't surprise me that
PQfnumber shows up so big in your profile. I think it would probably
make sense to introduce a PQfnumber variant that does not do the
downcasing/quote handling (called e.g. PQfnumberRaw).
However, I do think you could convert this per-row overhead in your
case to per-query overhead by caching the result of PQfnumber for each
unique C++ string_view. Afaict that should solve your performance
problem.