RFR:8146218: Producing streams in java.time?

Stephen Colebourne scolebourne at joda.org
Wed Jan 20 15:15:02 UTC 2016

I'm happy with the logic and specification of this proposal. I think it
will be a useful addition.

I'll let the Oracle team chime in to do a further review.


On 16 January 2016 at 13:31, Tagir F. Valeev <amaembo at gmail.com> wrote:

> Hello!
> Thanks for review! Here's the updated patch:
> http://cr.openjdk.java.net/~tvaleev/webrev/8146218/r2/
> SC> The docs say that if the end date is before the start date, the
> SC> stream is empty. While I can see just about why LongStream.range()
> SC> works that way, I don't think this API should. The minimum is an
> SC> exception, but it would be easy to support negative in the
> SC> days-only case or the months-only case. The problem is where there
> SC> are both months/years and days and those should be exceptions.
> Now datesUntil(endExclusive) throws an IllegalArgumentException if end
> date is before start date.
> datesUntil(endExclusive, step) supports negative periods. It throws
> IllegalArgumentException if:
> - step is zero
> - step.toTotalMonths() and step.getDays() both non-zero and have
> opposite sign
> - step is negative and end date is after start date
> - step is positive and end date is before start date
> Otherwise it works correctly: you can use
> LocalDate.of(2016, 1, 1)
>   .datesUntil(LocalDate.of(2015, 1, 1), Period.ofMonths(-1));
> Steps like Period.of(-1, -1, -1) are also supported.
> SC> The single-arg implementation uses plusDays() with an
> SC> incrementing number. When the performance patch goes in, the
> SC> proposed streaming implementation will be optimal for small
> SC> streams but less optimal for large ones. The fastest way to loop
> SC> over a list of dates would be to manually generate them by
> SC> incrementing the day until it exceeds the length of month, and so
> SC> on. However, this would be serial.
> As I understand, plusDays performance patch is already pushed.
> It's possible to write custom Spliterator which would use
> previous.plusDays(1) in tryAdvance() and jump (via
> LocalDate.of(startEpochDay+n)) in trySplit() if parallel processing is
> requested. However this adds at least one additional class and more
> complexity. I guess, such optimization can be considered later as
> separate issue when API is stabilized.
> SC> As such, I think the best way to write this, taking account of
> SC> how plusDays() is implemented, is as follows:
> SC> LongStream.range(start.toEpochDay(),
> SC> end.toEpochDay()).mapToObj(LocalDate::ofEpochDay);
> Ok, now it's done this way.
> SC> In the period-based method, it would be best to capture the case
> SC> where totalMonths == 0 and days > 0 and forward to another method
> SC> that ignores months. That private method can share implementation
> SC> with the public single-arg method (passing in 1).
> This case still more complex than one-day case as it requires division
> and multiplication. Thus I'd keep these case separately. However I
> simplified "months = 0" path using ofEpochDay, now it should be
> faster.
> SC> The docs for the period-based method should specify the strategy
> SC> that produces the results (in abstract terms). This needs to cover
> SC> that the result is equivalent to always adding to the start period
> SC> a multiple of the period.
> I added some clarifications, please check.
> SC> Some nits:
> SC> I prefer to avoid @link in the first sentence. Just using 'stream'
> would be sufficient.
> Done.
> SC> The first sentence should be a summary. In this case it probably has a
> bit too much detail.
> Done.
> SC> The @return has 'values' on a new line when it could be on the same
> line.
> I set now line length = 100 characters in my IDE. Is it ok?
> SC> If statements need braces.
> Done.
> With best regards,
> Tagir Valeev.

More information about the core-libs-dev mailing list