October 2010 - libdispatch-dev - lists.macosforge.org

Some questions about source handling
by cee1 11 Dec '10

11 Dec '10

Hi all, Questions: 1. How does a source's event handler accesses data returned by kevent? Data returned by kevent will finally be recorded is ds_data field in _dispatch_source_latch_and_call. The event handler will be called with ds_handler_ctxt, which shoule be do_ctxt. "do_ctxt" can be set by function dispatch_set_context. The problem is: - If a handler need to access ds_data, we have to set a proper do_ctxt (e.g. a do_ctxt of source). That's not convenient. - "ds_data" is a member of dispatch_source_t, and dispatch_source_t is an opaque struct to client code. 2. We have custom source of type DISPATCH_SOURCE_TYPE_DATA_ADD and DISPATCH_SOURCE_TYPE_DATA_OR, and can use dispatch_source_merge_data to fire custom sources. Why we don't have an 'is_level' source? An "is_level" source may not make much sense, but can be attached with cancel handler, this may be useful in some cases. -- Regards, - cee1

3 4

What's cooking for libdispatch in future?
by Danilo Egea 24 Oct '10

24 Oct '10

Some new feature is being developed for libdispatch in the near future? I don't see movement in subversion. For OSX Lion we will have something new? Thanks! sorry my english translated with help google :'( __________________________________________________ Fale com seus amigos de graça com o novo Yahoo! Messenger http://br.messenger.yahoo.com/

1 0

Windows Port
by Brent Fulgham 23 Oct '10

23 Oct '10

I've noticed that the WebKit builds on Windows have started making use of libdispatch. Do you think we might get access to these changes in the libdispatch project? I'd sure love to take advantage of them! Thanks, -Brent

3 2

global queue or many serial queues?
by Greg Pfeil 21 Oct '10

21 Oct '10

I'm converting some code to use libdispatch instead of my home-grown event queue, and I'm wondering what the best approach is. There is a tree, and each task in the queue operates on a node and potentially its parent and children. Before I converted my code to libdispatch, I had a single concurrent queue and each node in the tree had a mutex. When a node was operated on, it would lock its parent, itself, and its children in that order (this isn't optimal for the application, but it was a conservative approach that meant I didn't have to think about deadlocks). Obviously this can be translated pretty directly to libdispatch, using a global concurrent queue and semaphores. However, I was wondering if (even with all the semaphores) it might be beneficial to create a serial queue for each node. In my mental model, this would eliminate blocking on threads each trying to operate on the same node – the semaphores would really only cause blocking when two adjacent nodes are being operated on. Or is it the case that a single concurrent queue would work just as well – once a task is blocked, it just grabs another one and starts that. And, if I'm using multiple serial queues, is it also beneficial to suspend the queues for the parent and children nodes while the semaphores are locked? I know the answers depend on other questions like "how long does a task take to run?" (pretty short, fraction of a second), "how many queues will you have?" (can vary a lot, but "dozens" seems reasonable), "is the tree tall or wide?" (probably more wide), and probably plenty of other things. So, hopefully this will at least help me to understand the tradeoffs involved. Thanks.

2 1

Queue size
by Thomas Clement 07 Oct '10

07 Oct '10

> Queues cannot have variable widths - it's either 1 or "infinite". That idea was considered but ultimately dropped given that it only made submitting things to queues even less deterministic in terms of where/how they would run, and there is already a way (again, semaphores) to get that behavior, so this would have only added complexity to all queues for little overall gain. I'm confused. What about the dispatch_queue_set_width() private function? Isn't this already implemented and functional? Thomas

3 6