Hash synchronization patch, version 2

[Boehm, Hans]

It turns out that I already managed to expose another bug or two with some
more thorough multiprocessor testing.  I think they have very low occurrence
probability on a uniprocessor, but you might want to wait for version 3 ...

Hans

> -----Original Message-----
> From: Boehm, Hans [ mailto:hans_boehm@hp.com ]
> Sent: Wednesday, February 21, 2001 5:05 PM
> To: Java Discuss List
> Subject: Hash synchronization patch, version 2
> 
> 
> Here's a new version of the hash synchronization patch.  It 
> should fix a
> problem that Jeff ended up tracking down (incorrect thread identifiers
> resulting in illegal monitor state exceptions), another one that he
> mentioned (finalization cycles), as well as one other one 
> that I ended up
> running into when I finally managed to reproduce the first 
> problem (lack of
> locking in a finalizer needed to deallocate heavyweight locks).
> 
> I attached libjava/gcj/hash_sync.h separately, since I still 
> haven't found
> the cvs diff option to include it.
> 
> If this works for some other people, I would like to see it 
> checked into the
> development tree (not the 3.0 branch), but not yet enabled by 
> default (even
> on x86 linux, which is currently the only tested platform).  
> I doubt these
> were the last bugs in the patch, but there's only one way to 
> find the rest
> ...
> 
> Hans
> 
> 
>