<i18n dev> RFR: 8288723: Avoid redundant ConcurrentHashMap.get call in java.time [v2]

Roger Riggs rriggs at openjdk.org
Tue Jul 5 15:29:42 UTC 2022


On Sun, 3 Jul 2022 22:06:58 GMT, Attila Szegedi <attila at openjdk.org> wrote:

>> src/java.base/share/classes/java/time/format/DateTimeTextProvider.java line 312:
>> 
>>> 310:     private Object findStore(TemporalField field, Locale locale) {
>>> 311:         Entry<TemporalField, Locale> key = createEntry(field, locale);
>>> 312:         return CACHE.computeIfAbsent(key, e -> createStore(e.getKey(), e.getValue()));
>> 
>> If `createStore` can be static, the call site will only have to construct a constant lambda than having to pass `this` to construct a new instance on every call
>
> Well, if you _really_ want to noodle this for performance, you can also store a `this`-bound lambda in a `private final`  instance field, so then it's only created once too. I wouldn't put it past `javac` to do this, but I'd have to disassemble the bytecode of a little test program to see whether it's the case.

Can there be some JMH tests to confirm the performance?
The value domain of the keys is pretty limited (7 * 7) max; and I'm not sure that the combination of creating a new record and the hashcode and equals methods would be faster than string concat of a constant and a single digit integer.

-------------

PR: https://git.openjdk.org/jdk/pull/9208


More information about the i18n-dev mailing list