Re: libpq: Process buffered SSL read bytes to support records >8kB on async API

On Wed, Jul 16, 2025 at 11:50 AM Jacob Champion
<jacob.champion@enterprisedb.com> wrote:
> On Wed, Jul 16, 2025 at 11:11 AM Andres Freund <andres@anarazel.de> wrote:
> > Do you have a WIP patch?
>
> I'm working on one now.

The attached still needs some documentation work, and closer
inspection of the new assertions and OpenSSL error handling, but I
think it's good enough to get my intent across.

0001 is a fix for pqWait+gssencmode, which is hopefully the easier of
the two to review/backport. The current behavior for GSS is, IMO, an
obvious oversight.

0002 adds drain semantics at the end of pqReadData(). If we were
writing this from scratch, it wouldn't look like this: there's no
point in having the transport layer split data off into an undersized
conn->inBuffer if we're just going to come right back and drain it all
anyway. But I'm hoping this is easier to reason about and verify,
compared to a bigger rewrite.

I've renamed the existing pgtls_read_pending() to
pgtls_read_is_pending(), to try to disambiguate it from the new
*_drain_pending() APIs, but I'm not particularly attached to my choice
of names.

Thanks,
--Jacob