[Mono-list] Intrinsicly Non-Portable?

Christian Ratliff cratliff@lily.org
Fri, 23 Nov 2001 16:46:07 -0500


 
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello all,

  Has anyone given any thought to how we might deal with issues which
are intrinsically non-portable? By way of example, allow me to offer
two items.

System.Net.Sockets.Socket.Handle { get }
	This property getter returns a magical value "IntPtr". This value is
really a kernel event object gained by calling the Winsock function
::WSAEVENT ::WSACreateEvent(). This handle value is then attached to
a socket with the call ::WSAEventSelect(::SOCKET, ::WSAEVENT, long).
At that point you can nicely block against the ::WSAEVENT.
	No big deal yet, one can easily imagine using ::select() under Unix
to implement this, and having the socket handle be the magical
"IntPtr".

System.Threading.WaitHandle.WaitAny()
	This method exposes the functionality associated with the Win32
call:  ::WaitForMultipleObjectsEx(). This call allows a thread to
block against a wide variety of objects: semaphores, mutexes,
threads, events, and so on. If any or all of those objects move to
the signaled state, then the function returns. POSIX, in my
experience, has no analog.


  When you combine these two together, you wind up in a situation you
cannot easily replicate under POSIX+BSD sockets. You can use select()
to monitor a socket for events and pthread_cond_timedwait for the
condition variable (close and better than Win32 EVENTS). However,
there is not a pthread_mutex_timedlock for a timeout-able mutex lock
under Linux.
  Even if there was, the semantics would still not translate well. If
you do a System.Threading.WAitHandle.WaitForAny() with a 2ms timeout,
how do you "share" the timeout among the several different objects?
It is easy to imagine a situation where you start to divide below the
timing granularity.
  Needless to say, the .NET Framework betrays its parentage, and I
was wondering how people had planned to deal with this under the
various Unix platforms.

Christian

PS. For the record, Outlook wants to correct the word "IntPtr" with
"Inept". :)

+-----+
Christian Ratliff <cratliff@lily.org>
lily Development Group
"This is the very perfection of man, 
  to find out his own imperfections" -  St. Augustine



-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 7.0.3 for non-commercial use <http://www.pgp.com>

iQA/AwUBO/7DnrZBtyfwMNgGEQL8egCgvRQKhGUf2+JymOmMnCxFytYZ8CsAoJt4
eCRn/7ao+6dvNrOZR3o17oRk
=OZW5
-----END PGP SIGNATURE-----