Re: libseed-list handling gpointer args/returns

[Jonatan Liljedahl <lijon kymatica com>]

On 08/04/2010 03:21 AM, Alan Knowles wrote:
> On Wednesday, August 04, 2010 04:56 AM, Jonatan Liljedahl wrote:
> > Could you explain why simply setting arg->v_pointer = (gpointer) value
> > would break stuff? And why one should wrap JS values in GValues?
>
> The two use cases I can think of for the gpointer stuff are
> a) random data stored in C that can be recovered later by the Javscript
> program
> b) specific data that is needed by the C api (which has been lazily
> wrapped as void*)
>
> for a) when that data is sent back into Javascript, - eg. storing it on
> a tree element which can be retrieved later, there a has to be a way
> that the data can be recovered and turned back into a javascript object.
> for b) I guess the introspection should document it properly rather than
> pretending that its' void*

I agree on b). for a) it's just a matter of casting the js value to the 
gpointer, or do I miss something?

> > Regarding memory handling, when it's an arg to a function that uses
> > the pointer synchronously, then I guess nothing needs to be done. Just
> > pass the pointer.
>
> The problems you run into is, if the data you passed to the pointer
> get's freed by the Javascript engine before it get's used somewhere -
> this can be a problem where the call is async, or the data get's stored
> for later use.

Yes, but any API that takes a raw gpointer to be stored should (and most 
often does) have a callback for when the pointer isn't needed anymore. 
So, when we pass the values to the function, just JSValueProtect() them 
and cast to gpointer. Then in the destroy-callback, unprotect them.

The two cases that I'm especially looking for is g_async_queue_push/pop 
and g_thread_pool_push.

In the async_queue, there is no destroy-callback.. async_queue_push() 
would need to protect the value, and pop() should unprotect it. I don't 
know if this can be solved with gi (something with transfer-ownership?) 
or if this needs to be wrapped manually...

With the thread_pool, you would also need to handle the casting and 
unprotecting in the wrapper-function that calls the js function:

void thread_pool_wrapper (gpointer value, gpointer data) {
    SeedValue v = (SeedValue) value;
    (call data->func with v...)
    seed_value_unprotect(data->ctx, v);
}

but this might be a lot easier to do manually instead of using gi, so 
I'll put it in my dispatch module instead...

> > In other cases (like callback user_data) one would need to
> > JSValueProtect() it and unprotect it when it's not longer needed (when
> > the callback is destroyed). not sure how to solve that, but I guess
> > this is already done for other arg-types, perhaps depending on gir
> > info about ownership/scope/etc..?
>
> As your other email made me realize callback user_data is probably used
> internally to pass the closure stuff around. and Javascript data is
> kludged onto that..

Yes, also callback user_data is never needed on the script-side due to 
lexical scoping.

/Jonatan

> Regards
> Alan
>
> > /Jonatan
> >
> > On 07/29/2010 12:52 AM, Alan Knowles wrote:
> > > The whole pointer/ in out, memory handling needs quite some though.
> > >
> > > I've not come up with a idea for a solution that solves all, or even
> > > most scenarios without breaking stuff.
> > >
> > > It needs some trial and error probably
> > > my guess is that the solution might be only allowing GObjects to be
> > > stored in gpointer data - and wrap JS values up in GValues
> > >
> > > I've no idea how badly that might affect memory usage though...
> > >
> > > Regards
> > > Alan
> > >
> > >
> > >
> > > --- On 29/Jul/2010, Jonatan Liljedahl wrote:
> > > > many functions in GLib take gpointer as args or return values, for
> > > > example GLib.async_queue_push(queue, data).
> > > >
> > > > but this doesn't work currently, it gives errors like "ArgumentError
> > > > Unable to make argument 2 for function: async_queue_push."
> > > >
> > > > There is a hack in seed_gi_make_argument() that allows strings to be
> > > > passed, but this doesn't seem to work either (async_queue_pop() don't
> > > > seem to return a pushed string, but undefined)
> > > >
> > > > it should be possible and perhaps even simple to just wrap seed
> > > > objects/values in these cases.
> > > >
> > > > what function does the gvalue-to-seed conversion of the return value?
> > > >
> > > > /Jonatan
> > > > _______________________________________________
> > > > libseed-list mailing list
> > > > libseed-list gnome org
> > > > http://mail.gnome.org/mailman/listinfo/libseed-list
> > >
> > > _______________________________________________
> > > libseed-list mailing list
> > > libseed-list gnome org
> > > http://mail.gnome.org/mailman/listinfo/libseed-list
> >
> > _______________________________________________
> > libseed-list mailing list
> > libseed-list gnome org
> > http://mail.gnome.org/mailman/listinfo/libseed-list