Re: socket and timeouts

["Nic Ferrier" <nferrier at tapsellferrier dot co dot uk>]

>>> Bryce McKinlay <bryce@waitaki.otago.ac.nz> 28-Jun-01 5:05:50 AM
>>>

    I think it is a good plan. There are also some 
    other reasons why we should have separate 
    Socket IO from FileDescriptor, like implementing
    shutdownInput/shutdownOutput, and implementing 
    things correctly on native win32 where a socket is 
    not the same as a file descriptor.

You'll see my post to java@gcc.gnu.org about having done the impl.

What I've done is to add timeout support as per my original mail, I
haven't:

- added Windows timeout handling
- added intelligent shutdown handling

For the latter, I did program winsock about a million years ago (3.11
stuff) so I don't think I'm best qualified to add the code for that
(but I'll do it if no one else wants to).


But as far as intelligent shutdown handling goes I'm not entirely
sure what should be done. Should I be doing something to support
SO_LINGER? Presumably that's handled by glibc?

What I've done is:
- code the streams' close methods call the socket's close method.
- write a CNI close method which does a shutdown on the socket

The shutdown() is specified as:
- don't write any more data
- don't read any more data

Perhaps it should be different depending on the value of close()?

BTW: Classpath's impl just does a close() on the file descriptor.


Nic