Re: JNI crashes from IsSameObject in swig_connect_director

I went ahead and filed this as a bug: https://github.com/swig/swig/issues/1862

From: Holle, Jess
Sent: Friday, August 14, 2020 6:01 AM
To: [email protected]
Cc: Wynkoop, Mike <[email protected]>
Subject: JNI crashes from IsSameObject in swig_connect_director

I am seeing 2 different, strange JNI crashes from within JNI's IsSameObject in generated swig_connect_director methods on Android, i.e. from

  bool derived = (jenv->IsSameObject(baseclass, jcls) ? false : true);

In one case I get:

  JNI DETECTED ERROR IN APPLICATION: use of deleted local reference 0xd25

I can readily reproduce this issue by running while connected to Android Studio's memory profiler.

In the other case I'm getting reports of

  JNI ERROR (app bug): attempt to use stale Local 0x29 (should be 0x21)'

Yet 'baseclass' here has been found and a global reference created for it just above this and 'jcls' is passed down in the same JNIEnv on stack from the result of


  jenv->GetObjectClass(jself)

in a Java_*_1director_1connect call.

The only thing that's making any sense to me at this point is a race condition in the code:
    if (!baseclass) {
      baseclass = jenv->FindClass("com/ptc/daltonplayer/atomSDK/ActionDelegate");
      if (!baseclass) return;
      baseclass = (jclass) jenv->NewGlobalRef(baseclass);
    }

This code has a clear race condition - as it stuff `baseclass` into a static as a local reference and *then* turns around and replace it with a global reference.

I would like to propose a simple fix to this race condition:

    if (!baseclass) {
      jclass localBaseclass = jenv->FindClass("com/ptc/daltonplayer/atomSDK/ActionDelegate");
      if (!localBaseclass) return;
      baseclass = (jclass) jenv->NewGlobalRef(localBaseclass);
    }

--
Jess Holle

_______________________________________________
Swig-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/swig-user