In some cases, I would expect that the exception being overridden
would/should become the 'cause' of the new exception so it is not cleared
but chained. Does JNI support that?
On the original issue, discarding of exceptions should be explicit not
implicit.
Keep (or insert) the exceptionClear().
Roger
On 2/25/2014 6:35 AM, Chris Hegarty wrote:
On 25/02/14 11:26, Petr Pchelko wrote:
Hello, Alan.
I can see how this might be attractive but doesn't it mean you are
suppressing an important exception?
In case we’ve already got into the JNU_Throw we will throw a new
exception and override the original one anyway. However I agree that
this might suppress the warning for the code analysis tools.
Are the examples that you are looking at along these lines?
There are a number of examples when JNU_Throw is used to:
1. Replace the message of an original exception:
src/share/native/sun/awt/image/awt_ImageRep.c:192
There are a few such places.
2. Rethrow some meaningful exception if the call to some function
failed: src/windows/native/sun/windows/awt_Window.cpp:2861
This is a much more common use case. In this case we have a return
code from some method and we do not care if it was failed because of
JNI exception or for some other reason. This is the main case where
we need to add env->ExceptionClear() everywhere.
3. Quite common is to use it in the initIDs method to rethrow
NoSuchFieldError: src/share/native/sun/awt/image/BufImgSurfaceData.c:77
This one is questionable, I think that rethrowing is not needed here
at all, the original exception is much more informative.
Agreed. Similar to NetworkInterface.c Line:172
http://hg.openjdk.java.net/jdk9/dev/jdk/file/6ee5c47bdba7/src/solaris/native/java/net/NetworkInterface.c#172
-Chris.
4. Where currently are throwing an exception with pure JNI, but it
could be replaces with JNU:
src/windows/native/sun/windows/awt_PrintJob.cpp:1365
Similar to #2.
With best regards. Petr.
25 февр. 2014 г., в 2:42 после полудня, Alan Bateman
<alan.bate...@oracle.com> написал(а):
On 25/02/2014 10:31, Petr Pchelko wrote:
Hello, Core and AWT teams.
In AWT we have a lot of pending exception warnings which are now
being fixed. A big fraction of this warnings is about a pending JNI
exception at a call to JNU_Throw*.
Why don’t we add an env->ExceptionClear() call in the beginning of
each JNU_Throw function? It is absolutely safe because:
1. We are rethrowing an exception an loosing the original one anyway.
2. In case there’s no pending exception the ExceptionClear is a no-op.
If we do this the code would become much cleaner, because currently
we have to manually clear a pending exception before every call to
JNU_Throw*.
What do you think about this proposal?
I can see how this might be attractive but doesn't it mean you are
suppressing an important exception? We've fixed many areas in the
last few weeks and I think a common case was just that whoever wrote
the original code didn't realize that some JNI functions having a
pending exception when they fail. In those cases then often it is a
simple matter of just returning from the JNI function (maybe after
some clean-up). Are the examples that you are looking at along these
lines? I guess I'm mostly interested to know whether the JNU_Throw
usages are needed or not.
-Alan.
