memory allocation, finalization, jni code and locks (Was: [cp-patches] remove JamVM workarounds from GTK peers)

[Mark Wielaard]

Hi,

Moving the the main classpath list as a FYI for all runtime developers.
They probably know, but since we are now relying on this lets explicitly
say so.

On Wed, 2005-08-17 at 23:16 -0400, Thomas Fitzsimmons wrote:
> I'm removing these workarounds for a JamVM deadlock problem that was
> fixed in jamvm-1.3.2.
>
> -  /* Must release the GDK lock before allocating memory through the
> -     JVM, since some JVMs use the same lock for allocations and
> -     finalization.  Deadlock can occur on those JVMs. */
> -  gdk_threads_leave ();
> -
>    result_array = (*env)->NewIntArray (env, (width * height));
> -
> -  gdk_threads_enter ();

When memory allocation happens, even when called from JNI, finalization
should not happen on a thread that might hold a user visible lock, even
if it is a non-java lock. So garbage collection collection can occur on
any arbitrary thread, but running finalizers cannot.

Cheers,

Mark

signature.asc
Description: This is a digitally signed message part