Safepointing & Locking issue

[Keith McGuigan]

Nicolas Michael wrote:
> With DTrace I just saw -- thanks to the really cool hotspot probes (some of them
> seem to be new, it's the first time I see that great level of detail!) -- that
> there are about 1000 Exceptions occurring per second ("ExceptionOccurred-entry")
> (but less than one "ThrowNew-entry" per second). I'm not sure how ThrowNew
> relates to ExceptionOccurred and what they are actually counting, and whether
> 1000+ of ExceptionOccurred is "a lot" (it seems quite high ;-) ). Could
> exceptions also cause deoptimization?

The 'ExceptionOccurred-entry' and 'ThrowNew-entry' probes trace the 
corresponding JNI calls (from native code).  ExceptionOccurred is used 
to query whether an exception occurred or not, so seeing it called 1000 
times a second doesn't mean that there are 1000s of exceptions 
occurring.  Such frequent queries are probably not great for performance 
either, but that's a different story.

ThrowNew is called to actually throw an exception from native code, but 
that's not the only place where exceptions are thrown -- it's only when 
exceptions are thrown from native code using JNI.  And in any case I 
don't think throwing exceptions necessarily leads to deoptimization 
and/or safepoints.

--
- Keith