> On 14 Jul 2021, at 10:54, houzj(dot)fnst(at)fujitsu(dot)com wrote:
> Since PQfnumber() is not a cheap function, I think we'd better invoke
> PQfnumber() out of the loop like the attatched patch.
Looks good on a quick readthrough, and I didn't see any other similar codepaths
in pg_dump on top of what you've fixed.
> After applying this change, I can see about 8% performance gain in my test environment
> when dump table definitions which have many columns.
Out of curiosity, how many columns are "many columns"?
--
Daniel Gustafsson https://vmware.com/