RFR JDK-8187653: Lock in CoderResult.Cache becomes performance bottleneck
xueming.shen at oracle.com
Fri Mar 2 02:09:29 UTC 2018
On 3/1/18, 4:35 PM, David Holmes wrote:
> When you replace synchronized code with concurrent data structures you
> introduce race conditions that are precluded in the synchronized code.
> These need to be examined carefully to ensure they are safe. For
> example, whenever you replace a HashMap with a ConcurrentHashMap you
> need to see if put() needs to be replaced by putIfAbsent().
The assumption here is that putIfAbsent() does not help/save anything as
the value object
would have been created already when it reaches here. And it appears
there is no need here
to have the check&put to be atomic, replacing any existing key/value
pair is fine in this use
scenario. Was thinking about computIfAbsent(), but concluded it's just
little overdone (wonder
why we decided to use the WeakReference in this case, it probably is not
worth it, but keep it as
is for "compatibility" concern).
More information about the core-libs-dev