On Thu, Jan 9, 2025 at 8:17 AM Peter Eisentraut <pe...@eisentraut.org> wrote: > > Maybe it would work to just use plain "int" as the type here. Any > socket number must fit into int anyway in order for PQsocket() to be > able to return it. The way I understand Windows socket handles, this > should work.
Looks like it should work for current Windows, yeah. This is the approach taken by OpenSSL [1]. It'd be sad to copy-paste the API bug into a new place, though. If we're going to disconnect this API from SOCKET, can we use uintptr_t instead on Windows? If someone eventually adds an alternative to PQsocket(), as Tom suggested in [2], it'd be nice not to have to duplicate this callback too. --Jacob [1] https://docs.openssl.org/3.4/man3/SSL_set_fd/#notes [2] https://www.postgresql.org/message-id/153442.1624889951%40sss.pgh.pa.us
