DateTimeFormatter.format() uses exceptions for flow control

[David Holmes]

Hi Clement,

Please note that patches can only be accepted if they are sent through, 
or hosted upon OpenJDK infrastructure. If the patch is small enough can 
you send it inline in the email (attachments are often stripped) 
otherwise you will need to find an OpenJDK Author who can host the patch 
for you on cr.openjdk.java.net

Thanks,
David

On 10/10/2016 1:24 AM, Clément MATHIEU wrote:
> Hi !
>
> I noticed that DateTimePrintContext.getValue() relies on exceptions to
> handle optionality. Using exceptions for flow control seems both
> unexpected and very costly, ie. I discovered the issue
> when  LocaleDate.format(BASIC_ISO_DATE) showed up as a heavy hitter in
> my application.
>
> Formatting a LocateDate as a "yyyyMMdd" string should take ~0.1μs but
> currently takes from ~2.5μs, stack depth = 0, to ~10μs, stack depth =
> 128 when an optional parser is defined but the optional field is not
> supported. This seems unfortunate when exceptions can easily be avoided
> by testing if the field is supported before trying to get its value.
>
> Webrev:
>
>   http://cdn.unportant.info/openjdk/dtf_exceptions/webrev.00/
>
> JMH benchmarks:
>
>   https://gist.github.com/cykl/020e1527546a1ba44b4bb3af6dc0484c
>
>
> What do you think ?
>
>
> Note: Many tests within java.time are currently broken because of
> TestNG upgrade, see https://bugs.openjdk.java.net/browse/JDK-8152944.
> Would it help if I spend some time adding the missing L suffixes ?
>
> Regards,
>
> Clément MATHIEU
>