[vlc-devel] [PATCH] libvlc: add API to propagate mouse move and down/up events

[Rémi Denis-Courmont]

Le mardi 6 mai 2014, 17:45:26 Felix Paul Kühne a écrit :
> >> You don't know which vmem instance you are referring to. This is totally
> >> broken if you have more than one window.
> 
> This is absolutely true.
> 
> However: libvlc_video_get_cursor() already uses the same pattern I do in
> this patch. It lets you access the video output by N.

That is a known design bug that should not be repeated. I don't remember why 
get_cursor() was even needed way back; it seems useless/unusable in practice 
or almost. But at least it exhibits dubious correctness in returning the 
correct coordinates for _the_last_ focused video area.

> Additionally, if you consider the remainder of the video API listed in
> libvlc_medoa_player.h, it appears to me as if more than one video output
> per libvlc instance was never a supported scenario.

You've got to be kidding. libvlc_video_get_size() supports it (for meta data) 
The vmem API supports it too; there was even a discussion a few months back on 
how to add new per-vmem instance functions.

Also several video functions apply correctly to all video outputs, regardless 
of how many there are.

(...)
> I fully agree, however, this is not really feasible in the scenario targeted
> by the patch.

vmem is neither designed nor capable of supporting video rendering (it was 
designed by Sam to feed video post-processing filters). Mouse handling is not 
the (worst) limitation inhibiting proper rendering support. Adding a function 
pretending the contrary is misleading to external developers and plain wrong.

> For the NPAPI plugin, we support a special rendering mode called
> “windowless”, which does not make use of the existing libvlc vout modules.

That is a bug in the NPAPI plugin and a gap in modules/video_output/, not a 
problem in LibVLC at this point. Using vmem as it currently stands cannot work 
correctly for that scenario, regardless of this patch.

-- 
Rémi Denis-Courmont
http://www.remlab.net/