How select() works

First I know it's not a good idea to non-block a socket.

Second, how can a non-blocking connect() work, i'm not writing a program for a lan network with replies under 1ms... The times over the internet are pretty high, if i non-block connect() it will simply never connect!

Ok, you've all given me what select() does. I'm pretty comfortable with monitoring file descriptors for reading: whenever someone writes on a socket, select returns, and you can check what socket was written to and then perform a read. But I still cannot think of an example of when I would monitor sockets for writing. All the code I've seen on the Web monitors sockets for reading, so I would very much appreciate it if someone could give me an example along with code of when I would monitor sockets for writing.

Thanks,
Yasir

Try sending a large file somewhere, and see how long each successive call to send() takes.

Before long, you'll see send() taking a lot longer than it used to (because all the interal buffers are full), and you have to wait around for them to clear.

Originally Posted by Devil Panther

First I know it's not a good idea to non-block a socket.

Second, how can a non-blocking connect() work, i'm not writing a program for a lan network with replies under 1ms... The times over the internet are pretty high, if i non-block connect() it will simply never connect!

anyway is there a way to change the default timeout value of connect(), without changing the socket to non-blocking...

Originally Posted by Salem

Before long, you'll see send() taking a lot longer than it used to (because all the interal buffers are full), and you have to wait around for them to clear.

Ok, so monitoring write file descriptors are used to see when the internal buffers are empty so you can start writing again?

Originally Posted by Devil Panther

anyway is there a way to change the default timeout value of connect(), without changing the socket to non-blocking...

So want to time out if you can't connect() to a server, right? You could set an alarm(2) and then after the alarm goes off, handle it through signal(2). I've used this approach a long time ago in the client process to see if the server has gone down while the client is waiting for a response. I can't remember how I did it, though.

Check out http://www.cs.utk.edu/~plank/plank/c...p/lecture.html and http://www.cs.utk.edu/~plank/plank/c...s/Setjmp/sh4.c

Yasir

As I was implementing what I was suggested to you, I found out that connect() returns immediately when there isn't someone to connect to. I thought it would wait. alarm and signal won't work then.

the problem that it doesn't return immediately!
infact i have to wait a whole minute for it to timeout, and it's killing me.
which is the default timeout value for this function.

connect() doesn't return immediately? It does on my machine. But if that's what's happening on your machine, use the signal() alarm() combination I suggested and look at the links I posted.

are you saying i can interupt the connectcall in the middle with that???

I think so. Try it. Set up a signal with SIG_ALARM and call alarm with some number of seconds. Set up your setjmp calls, and in the signal function handler use longjmp() to jump back to the place you called setjmp(). The key is that setjmp initially returns 0, but when you call longjmp(), it returns what you passed in longjmp()'s second parameter.

hmmm since connect blocks, the program "stops" on that call to connect. I'm guessting i need to use more than one thread / process to do that, no?

What don't you like about my solution? Have you tried it? I would like to try it, but on my machine, connect() returns immediately when there isn't a server to connect to.

Timing out connect():
unix network programming