[FFmpeg-devel] [PATCH 1/2] lavc: add JNI support

[Matthieu Bouron]

On Tue, Mar 01, 2016 at 08:13:48PM +0100, wm4 wrote:
> On Tue, 1 Mar 2016 20:10:50 +0100
> Matthieu Bouron <matthieu.bouron at gmail.com> wrote:
> 
> > On Tue, Mar 01, 2016 at 07:56:30PM +0100, wm4 wrote:
> > > On Tue, 1 Mar 2016 19:52:08 +0100
> > > Matthieu Bouron <matthieu.bouron at gmail.com> wrote:
> > >   
> > > > On Fri, Feb 26, 2016 at 05:36:40PM +0100, Matthieu Bouron wrote:  
> > > > > On Fri, Feb 26, 2016 at 4:41 PM, Matthieu Bouron <matthieu.bouron at gmail.com>
> > > > > wrote:
> > > > >     
> > > > > > On Mon, Feb 22, 2016 at 12:20:35PM +0100, Matthieu Bouron wrote:    
> > > > > > > From: Matthieu Bouron <matthieu.bouron at stupeflix.com>
> > > > > > >    
> > > > > >    
> > > > > 
> > > > > [...]
> > > > > 
> > > > >     
> > > > > >
> > > > > > Patch updated with the following differences:
> > > > > >   * fix of switch/case code style
> > > > > >   * removal of a miss declaration of FFJNIField enum as a global variable
> > > > > >
> > > > > >    
> > > > > Patch updated with the following differences:
> > > > >   * fixes a few typo in comments
> > > > >   * fixes a if statement in ff_jni_init_jfields    
> > > > 
> > > > Patch updated with the following differences:
> > > >   * fixes a few typo in comments and message logs
> > > >   * add missing .so at end of library names when trying to find
> > > >   JNI_GetCreatedVMs symbol (also add libart.so)
> > > >   * reintroduce public functions that allow the user to set the Java
> > > >   virtual machine (av_jni_(set,get)_java_vm) as the private C++
> > > >   JniInvocation wrapper is not available on all devices (android >= 4.4)  
> > > 
> > > I'm wondering if the VM should be stored per AVCodecContext. (Since it
> > > has to be set explicitly again by the user.)  
> > 
> > I think it is fine to store one VM for all the AVCodecContext as it will
> > be the same during all the lifetime of the application. I should also
> > enforce that the VM cannot be changed afterwards.
> > 
> > av_jni_set_java_vm stores the Java VM pointer locally in jni.c and not in
> > ffjni.c and it is retrieved in jni.c using av_jni_get_java_vm. I've done
> > it that way because if at some point we are to include ffjni.c from
> > libavformat it won't work (it will only set the java vm in the libavcodec
> > memory space).
> 
> The problem is that this is not library-safe. What if two libs within
> the process (which both are using libavcodec) are setting a different
> VM?

In the Android application context, you only have, AFAIK, one VM running
during the lifetime of the application. The code does handle the general
JNI case (even outside Android) but do we really want to do that anyway ?