Re: [AD] user-defined events

[Evert Glebbeek <eglebbk@xxxxxxxxxx>]

On 26 Sep 2008, at 20:58, Peter Wang wrote:
> Here is a proposal for allowing user-defined events.

Looks ok to me.

> Do we need an ALLEGRO_USER_EVENT_SOURCE?

I don't know. What would be the benefit? Maybe it'd just be a  
cosmetic thing?

> Aside: I'm considering simplifying some of the internals (again) by
> allocating events inside event *queues* instead of event sources.   
> So an
> event source registered with two queues would create two copies of  
> each
> event, one in each queue.  The upside is that we get rid of some ugly
> reference counting, which is only needed for such a rare situation.

I actually do that in the program I'm working on: when exiting it  
pops up a window asking whether you want to quit or not. That window  
uses its own event queue. I think I only did that because I was  
playing around anyway, I could have just reused the original queue.  
If it makes the code simpler, go for it!

> You wouldn't need al_allocate_user_event() then. Instead, you'd  
> allocate
> an event on the stack and fill it in.  It would be copied into each
> event queue when you call al_emit_user_event().

I think that makes sense, actually. However, it does sound like it  
might require a lot of (fairly simple) changes throughout the library?

> +/* XXX Currently we have no formal way to allocate event type  
> numbers. */
> +#define MY_EVENT_TYPE   1025

True, and we probably want one. Even if we define  
ALLEGRO_EVENT_ABOVE_HERE_FREE_FOR_USERS (or something like that),  
it's possible addons will want to use a range of event types as well.

How about the following: we make a function where a user can request  
the free "base number" for a range of events (say, 10), eg,
int my_first_event = al_get_event_range(10);
The user can then freely use numbers in the range my_first_event+0  
through my_first_event+9. The only downside is that users can't use a  
switch statement to identify their events.

> +typedef struct ALLEGRO_USER_EVENT
> +{
> +   _AL_EVENT_HEADER(struct ALLEGRO_EVENT_SOURCE);
> +   void *data1;
> +   void *data2;
> +   void *data3;
> +   void *data4;
> +} ALLEGRO_USER_EVENT;

I know I said providing the user with one pointer would be fine, so I  
can't reasonably complain when you give them 4, but I do think the  
common case might be that you want one or two integers to pass  
informmation along. Maybe we can have two of those as well?

Evert