[vlc-devel] [PATCH] access: add rfb/vnc viewonly module.

[Rémi Denis-Courmont]

Le dimanche 24 février 2013 21:38:30, Francois Cartegnie a écrit :
> Le 21/02/2013 19:32, Rémi Denis-Courmont a écrit :
> >> The fake+8bpp remoteosd filter trick being not sufficient and
> 
> accessible to
> 
> >> average user.
> > 
> > Yes. Please remove it in a separate patch.
> 
> Can't: Not intended for the same use.

I don't see the implication.

> >> +#define DEMUX_TAG (void *)0x1
> > 
> > This is wrong and will cause a warning on 64-bits platforms.
> 
> Wants a (void *) tag.
> "makes pointer from integer without a cast"

Then WTF did you submit the patch? After several years, you should know better 
than to submit evidently broken patches.

> >> +
> >> +    float f_fps;
> >> +    int i_frame_interval;
> >> +    mtime_t i_starttime;
> > 
> > This is not needed. Worse, it will prevent synchronizing with live
> 
> capture
> 
> > slave inputs.
> 
> Still needed for computing DEMUX_GET_TIME.

I don't see  that.

> >> +    return var_InheritString( p_demux, CFG_PREFIX "password"
> > 
> > Looks leaky.
> 
> Managed by libvnc.
> 
> > For the umpteenth time. Please don't do arbitrary wakeups.
> > 
> > If you need to sleep in an access_demux, use your own thread (e.g. V4L2).
> 
> I assume then that demux_t->pf_demux() isn't called its own separate
> thread.

pf_demux is called in the input thread. Seriously, you could read the code, 
and use gdb if in doubt. You are not a newbie anymore.

> >> +    p_sys->p_client->MallocFrameBuffer = mallocFrameBufferHandler;
> > 
> > Most probably the whole point of this callback is to avoid raw picture
> 
> memory
> 
> > copying. Please avoid memory copying then.
> 
> Problem is the framebuffer is built incrementally.
> I've change the way it is handled, but ended with a block_Duplicate...

Well, if it's used as a reference frame, at demuxer level, then you are forced 
to add one more memory copy.

> Thought also about overriding block pf_release to use a unique or dual
> fb, but it wouldn't be able to synchronize writes and caching would
> provide stalled screen.

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