[Webkit-unassigned] [Bug 61287] [UNIX] Don't use SOCK_DGRAM in socketpair()

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=61287





--- Comment #24 from Carlos Garcia Campos <cgarcia at igalia.com>  2011-05-26 05:27:42 PST ---
(In reply to comment #23)
> First you say STREAM is not used..

Yes, I meant it's used only as a fallback for non-linux systems.

> > You could also read more than one message with a single recvmsg() reducing the amount of syscalls. 
> 
> ..then you argument justification of the complexity due to a STREAM feature? 

you are right, that comment is  out of place.

> Look, I gotta stop right now wasting your and my own time. I'm not a reviewer, I don't have much to say about this issue and you don't need to listen to my argumentation. Certainly this is even more the case with Gtk port.

I appreciate your argumentation, the Qt port will have the same problem when plugin process is implemented, I'm just trying to find the better way to fix it.

> 
> To conclude my stand here: You are solving the problem of notifying of closing the connection. I don't think implementing entirely correct, arbitrary stream protocol parsing is the way to do it. The IPC is designed as a datagram protocol. Requiring 100% correct stream protocol implementation is insanity.

I agree, but this is only for the fallback case, what I was trying to say is that the same implementation that works for stream sockets, works for dgram sockets too without performance penalty. 

> I could think of few other, far simpler implementations for close notifications. One is the socket thing, which solves the crash problem too.

I don't think the socket thing is that easy, we would need another pair of connected sockets, right? so we would need to keep two sockets per connection and close both when the connection finishes. But only when SOCK_SEQPACKET is not available.

> "The orderly shutdown" scenario can be solved by also sending a simple message. For example, you can send datagram "KIMMO" to signify close, and that will work.

That was my first proposal, and apple guys said: there must be another way to do it.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.