Re: [Vala] "vala friendly" atomic pointer

[Maciej Piechotka <uzytkownik2 gmail com>]

On Thu, 2013-12-19 at 17:16 +0200, Tal Hadad wrote:
http://www.valadoc.org/#!api=gobject-2.0/GLib.WeakRef
> This is a simple solution thanks to the good design of GObject.
>
> Tal
That might not always work, unfortunatly, due to how structs are
implemented[1]. Beside - they are implemented currently using global
lock in GObject[2].

> From: norjaidi tuah ubd edu bn
> > To: vala-list gnome org
> > Date: Thu, 19 Dec 2013 08:59:44 +0800
> > Subject: [Vala] "vala friendly" atomic pointer
> >
> > Dear all,
> >
> > I want to implement something like these:
> >    *  x = AtomicPointer  get (ref atomicRef)   // reader
> >
> >    *  AtomicPointer  set (ref atomicRef, newref)  // writer
> >
> >    *  return AtomicPointer  get (ref atomicRef)  // reader
> >
> > but have the benefit of vala ref counting.
> >
> > The best I can think of is:
> >
> >    *  lock (atomicRef) { x = atomicRef; }
> >
> >    *  lock (atomicRef) { atomicRef = newref; }
> >
> >    *  lock (atomicRef) { return atomicRef; }
> >
> > But lock is too course.
> > e.g., lock (atomicRef) {x = atomicRef;} will also lock
> > out other readers, whereas AtomicPointer.get won't.
> >
> > I realise that the exclusive region is small.
> > But if you have a lot of readers, the effect
> > may be significant.
> >
> > Any suggestions?
> >
> >
> > Nice day
> > Nor Jaidi Tuah
The combination of lock free and reference pointers is 'hard' (as long
as you don't have double word CAS operation which is not implemented on
practically any hardware). The only way I know about is by
HazardPointers.

Gee 0.8 or newer have it's implementation of HazardPointers[3]. The
basic usage is as follows:

MyObject *object_ptr;

MyObject *read_object = HazardPointer.get_pointer<MyObject>
(&object_ptr);

The drawback is that you need to ensure that all operations are in
HazardPointer.Context. You want the context to live in thread as long as
you know that operation is taking place. Upon destruction all pointers
are freed (exact method is specified by policy).

Regards

[1] https://bugzilla.gnome.org/show_bug.cgi?id=703996
[2] https://bugzilla.gnome.org/show_bug.cgi?id=705728
[3] http://www.valadoc.org/#!api=gee-0.8/Gee.HazardPointer