|
I’ve got an existing windows based project which uses the base PJSUA API (from release 2.4.5). I’m trying to build an android app using this api. I’ve followed the Getting-Started/Android instructions to build libpjsua2.so without video support. The framework I’m using for app development allows building native android extensions (rather than using java) so I’ve grabbed libpjsua2.so and the headers from pjlib, pjlib-util, pjmedia, pjnath and pjsip folders. I’ve built an extension framework using this which seems to be functional, I’m able to make/receive calls and register with a registrar. I’ve now set up a longer term test. Essentially a secondary device is calling my android device and hanging up, continuously. I’m finding that once the android app receives ~250 calls it crashes with the following: 06-21 11:08:53.350 4459 4484 W dalvikvm: JNI local reference table (0x627fa7b0) dump: 06-21 11:08:53.350 4459 4484 W dalvikvm: Last 10 entries (of 512): 06-21 11:08:53.350 4459 4484 W dalvikvm: 511: 0x418044c0 int[] (1 elements) 06-21 11:08:53.350 4459 4484 W dalvikvm: 510: 0x418044d8 java.lang.ref.WeakReference 06-21 11:08:53.350 4459 4484 W dalvikvm: 509: 0x41804420 android.media.AudioTrack 06-21 11:08:53.350 4459 4484 W dalvikvm: 508: 0x41804420 android.media.AudioTrack 06-21 11:08:53.350 4459 4484 W dalvikvm: 507: 0x41804350 android.media.AudioRecord 06-21 11:08:53.350 4459 4484 W dalvikvm: 506: 0x418f22e8 android.media.AudioTrack 06-21 11:08:53.350 4459 4484 W dalvikvm: 505: 0x418f2228 android.media.AudioRecord 06-21 11:08:53.350 4459 4484 W dalvikvm: 504: 0x419695e0 android.media.AudioTrack 06-21 11:08:53.350 4459 4484 W dalvikvm: 503: 0x41969510 android.media.AudioRecord 06-21 11:08:53.350 4459 4484 W dalvikvm: 502: 0x41997038 android.media.AudioTrack 06-21 11:08:53.350 4459 4484 W dalvikvm: Summary: 06-21 11:08:53.350 4459 4484 W dalvikvm: 6 of java.lang.String (6 unique instances) 06-21 11:08:53.350 4459 4484 W dalvikvm: 1 of int[] (1 elements) 06-21 11:08:53.350 4459 4484 W dalvikvm: 1 of java.lang.ref.WeakReference 06-21 11:08:53.350 4459 4484 W dalvikvm: 251 of android.media.AudioRecord (251 unique instances) 06-21 11:08:53.350 4459 4484 W dalvikvm: 252 of android.media.AudioTrack (251 unique instances) 06-21 11:08:53.350 4459 4484 W dalvikvm: 1 of com.ideaworks3d.marmalade.LoaderThread 06-21 11:08:53.350 4459 4484 E dalvikvm: Failed adding to JNI local ref table (has 512 entries) 06-21 11:08:53.350 4459 4484 I dalvikvm: "Thread-290" prio=5 tid=11 RUNNABLE 06-21 11:08:53.350 4459 4484 I dalvikvm: | group="main" sCount=0 dsCount=0 obj=0x410335b8 self=0x59fd8690 06-21 11:08:53.350 4459 4484 I dalvikvm: | sysTid=4484 nice=0 sched=0/0 cgrp=apps handle=1528506816 06-21 11:08:53.350 4459 4484 I dalvikvm: | state=R schedstat=( 0 0 0 ) utm=91469 stm=30801 core=1 06-21 11:08:53.350 4459 4484 I dalvikvm: at android.media.AudioTrack.native_setup(Native Method) 06-21 11:08:53.350 4459 4484 I dalvikvm: at android.media.AudioTrack.<init>(AudioTrack.java:326) 06-21 11:08:53.350 4459 4484 I dalvikvm: at android.media.AudioTrack.<init>(AudioTrack.java:267) 06-21 11:08:53.350 4459 4484 I dalvikvm: at com.ideaworks3d.marmalade.LoaderThread.runNative(Native Method) 06-21 11:08:53.350 4459 4484 I dalvikvm: at com.ideaworks3d.marmalade.LoaderThread.run(LoaderThread.java:1137) 06-21 11:08:53.350 4459 4484 I dalvikvm: 06-21 11:08:53.350 4459 4484 E dalvikvm: VM aborting 06-21 11:08:53.350 4459 4484 F libc : Fatal signal 11 (SIGSEGV) at 0xdeadd00d (code=1), thread 4484 (Thread-290) Looking at the pjsip library sources in android_jni_dev.c, I’ve established the following: Whenever the record and track class+object references are obtained, ‘Creating Android JNI stream’ is logged. Whenever the record classes+object references are released ‘Audio record released’ is logged. Whenever the track classes+object references are released ‘Audio track released’ is logged. Due to this, I’ve checked the logs over a time period where the android application crashed, verifying that the number of times ‘Audio record released’ is logged matches the number of times ‘Creating Android JNI stream’ is logged (also
checked for track as well). Since these match, I’m confused as to what’s going on. It seems like the references are being released in c++ but not being released in java. The only way I can imagine this could happen is if the release from c++ doesn’t actually release
the reference but marks it for later removal and some background process which is meant to do the clean-up isn’t happening. I’m currently looking through the java-swig pjsip interface to see if I can find anything which may resolve this issue, but I’m not entirely sure what I’m looking for. Does anyone know how to fix this/any other ways to approach this than staring at the java-swig interface? |
_______________________________________________ Visit our blog: http://blog.pjsip.org pjsip mailing list pjsip@xxxxxxxxxxxxxxx http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
