Re: [Xen-API] [Xen-devel] [PATCH 10 of 15] libxc/ocaml: Add simple binding for xentoollog (output only)

[Rob Hoes <Rob.Hoes@xxxxxxxxxx>]

> > > Another idea might be to make the bindings use the async interfaces
> > > wherever possible by default, by definition anything potentially
> > > blocking has supports this and that would avoid the need for
> > > enter/leave, but at the expense of making the ocaml callers ugly
> > > perhaps? Or maybe this sort of thing ends up looking very natural in
> > > ocaml? Depends on your application's event mechanism I suspect.
> > >
> > > Last half witted idea: everything could be async but the bindings
> > > include the loop to wait for the async event, i.e. effectively
> > > making the call sync again. This sounds silly but it might allow
> > > better control over the placement of enter/leave vs callbacks, since
> > > you would just drop it over libxl_event_wait?
> >
> > Making everything explicitly async and non-blocking is by far the
> > preferred solution in terms of stability, as it largely removes the
> > need to worry about the GC interface and thread interactions.
> >
> > There are several libraries to wrap async interfaces in convenient
> > synchronous programming, most notably Lwt (http://ocsigen.org/lwt).
> 
> I'll have to take your word for that ;-)
> 
> > With this interface, all the OCaml callbacks are implemented in OCaml,
> > and it just needs a select/epoll or equivalent to wake up sleeping
> > threads when an IO event occurs.
> 
> On the libxl side you'd want to be using the ao_how thing described in
> libxl.h[0] and probably the event loop stuff in libxl_event.h[1]. My gut
> feeling is that you'd want to go the libxl_osevent_register_hooks route to
> integrate libxl into lwt's event loop rather than the
> libxl_osevent_{before,after}poll option.
> 
> [0] http://xenbits.xen.org/hg/xen-
> unstable.hg/file/1c69c938f641/tools/libxl/libxl.h#l377
> [1] http://xenbits.xen.org/hg/xen-
> unstable.hg/file/1c69c938f641/tools/libxl/libxl_event.h

Great, I completely forgot about the possibility of doing async calls to libxl. 
This sounds like an ideal solution, especially if we can avoid messing with the 
GC system this way (those segfaults are really hard to debug). It may be a nice 
excuse to get started with Lwt as well ;)

I'll do some investigation on how to integrate this.

Cheers,
Rob

> 
> Ian.
> 
> >
> > Dave has already done some bindings to Lwt, so I've CCed him...
> >
> > -anil
> >
> 

_______________________________________________
Xen-api mailing list
Xen-api@xxxxxxxxxxxxx
http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api