apr-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Saxon Druce" <sa...@blocksoftware.com>
Subject Re: [PATCH] WIN32 implementation of apr_poll_revents_get()
Date Wed, 05 Jun 2002 04:04:54 GMT
Hi,

----- Original Message -----
From: "JF B" <jfbriere@yahoo.com>
Sent: Wednesday, June 05, 2002 2:54 AM

> What is exactly the problem with the getsockopt() function?

Nothing that I know of. I've seen posts on other issues to this list talk
about reducing the number of system calls, so I thought maybe the extra call
to getsockopt() was bothering some people.

> My implementation is slightly different from yours. I use the getsockopt()
> function only when WSAGetLastError() returns WSAENOTCONN.

Yeah, I wasn't saying mine was better or anything ;) I'd planned to repost
my patch on its 2 month anniversary, but I was away for that week, so I was
going to when I got back. Then your post came in, so I thought I'd throw my
old patch into the mix as well. Yours is probably actually better since it
doesn't do the getsockopt() for the probably more usual case of an
already-connected socket. However, my patch also includes something to make
zero bytes received result in a hangup event. Maybe a combination of the two
patches would be best overall.

> By the way, why in your implementation don't you test the return of
> getsockopt()?

Good question, I probably should :)

> Also, I considered first using a flag in the apr_socket_t structure, but
my
> problem is that there is already one: 'disconnected'. I would use this
> flag, but after many review of the code, I still don't know the utility of
> this flag.

I don't know either, but it looks like it's used for something else. I was
considering a different flag called 'connected', like the unix code uses if
poll() is being emulated with select(). It's set to true in apr_accept() and
apr_connect(), and then apr_poll_revents_get() checks it before trying to do
the recv().

btw, if you're using apr_poll() with more than one socket in the poll set,
after the method returns only the sockets which were actually signaled will
be left in the set (in win32). I posted this problem nearly two months ago
to the list, asking for confirmation that this was wrong, and offering to
write a patch to fix it if so. I was going to wait until my first patch to
poll.c was committed before chasing this up any further, though.

bye,
Saxon Druce (saxon@blocksoftware.com)



Mime
View raw message