[concurrency-interest] The parts of the ReentrantReadWriteLock hold no reference to it
dl at cs.oswego.edu
Tue Oct 17 19:53:12 EDT 2017
On 10/17/2017 05:02 PM, Martin Grajcar wrote:
> The parts of the ReentrantReadWriteLock holding no reference to it may
> lead to a situation where the ReentrantReadWriteLock gets GC'ed while
> its read and write locks are still in use. This is alright unless
> the ReentrantReadWriteLock is weakly cached as e.g., in Guava Striped,
> where it has lead to the bug https://github.com/google/guava/issues/2477
> I'd suggest to add a reference to the enclosing class, e.g., by removing
> the static modifier. As all these objects are lightweight and not meant
> to have millions of instances, the cost is negligible. The advantage is
Some program out there probably has more ReentrantReadWriteLocks than
any of use can imagine, so will be hurt by this.
Still, I suppose we have done worse things to help solve
not-our-problem problems in even more crazy uses than yours,
so we probably ought to do it.
Maybe the added overhead will drive more people to use StampedLocks,
as we keep hoping.
> decreased chance of buggy user code and avoiding lengthy workarounds
> like https://github.com/google/guava/commit/957c1a5455508120d224f6d0d8f3bf8afa3630f0
> The cost benefit ratio seems to be good.
> Concurrency-interest mailing list
> Concurrency-interest at cs.oswego.edu
More information about the Concurrency-interest