I wrote:
> While looking over the thread, I remembered I wanted to convert
> strerror_r into a wrapper as well. Think I'll go do that next,
> because really it'd be better for snprintf.c to be calling strerror_r
> not strerror.
So while chasing that, I realized that libpq contains its own version
of the backend's win32_socket_strerror code, in libpq/win32.c.
This probably explains why we've not heard complaints about bogus
socket error reports from libpq; it's that code that's handling it.
What I think ought to happen is to merge win32.c's version of that
code into strerror.c, which'd allow removing win32.c and win32.h
altogether. However, not having a Windows environment, I can't
test such changes and probably shouldn't be the one to take point
on making the change. Anybody?
regards, tom lane