RFR(S): 8055004: Reduce allocation overhead in java.time.Period/Duration parse methods

[Claes Redestad]

On 2014-08-13 21:49, Xueming Shen wrote:
> looks fine.
Thanks for reviewing!
>
> though arguably it might be more correct to do
>
> if (dayStart != -1 || hourStart != -1 || minuteStart != -1 || 
> secondStart != -1) {
> ...
> }
>
> as the regex spec says the start/end return -1 if no match for the group.

start(int)[1] will return either the start index of the group (0 <= 
index < text.length(), implying index >= 0) or -1,
so I could argue both ways are fine, no?

Thanks!

/Claes

[1] 
http://docs.oracle.com/javase/8/docs/api/java/util/regex/Matcher.html#start-int-

>
> -Sherman
>
>
> On 08/13/2014 05:50 AM, Claes Redestad wrote:
>> Hi,
>>
>> can I have a review of this performance improvement to 
>> java.time.Period#parse and java.time.Duration#parse?
>>
>> bug: https://bugs.openjdk.java.net/browse/JDK-8055004
>> webrev: http://cr.openjdk.java.net/~redestad/8055004/webrev.0
>>
>> The patch avoids String allocation through matcher.group(n) calls, 
>> utilizes recently added
>> capability of parsing CharSequence offsets and improves on the rather 
>> awkward way fractions of a second
>> was aligned to nanosecond precision in java.time.Duration.
>>
>> Microbenchmarks show a 1.2-1.5x throughput improvement in both cases.
>>
>> Thanks!
>>
>> /Claes
>