Sorry Yoann, it was a crazy week and I missed this reply somehow...
Yes, you did an awesome job with setting up tests. We will obviously make
sure those continue to work as we move forward.
I did mention in the original email (I think anyway, I meant to) that there
seemed to be a problem in parsing with Java 8 in certain scenarios. I
found this via a SO post asking why something was not working the way I
expected. The answer was that there is a bug in Java 8, but that parseBest
seems to work consistently across both versions. We plan to talk about
moving to Java 9 as a base anyway so this may not be an issue either way we
decide to go regarding parse or parseBest.
On Wed, Jan 8, 2020 at 4:21 AM Yoann Rodiere <yoann(a)hibernate.org> wrote:
> This does nothing with Type. The way the grammar is defined it
literally understands each piece of the temporal. So given, e.g.,
{2020-01-01}, we know that 2020 is the year, etc. This is the benefit of
defining it syntactically.
I trust you can build a temporal correctly from a string. I'm more
concerned about passing that information to the JDBC driver through a
parameter, or even directly to the database through an SQL literal. Last
time I checked you had to use java.sql types to pass temporal parameters to
JDBC drivers, so you will have to convert the java.time value to a
java.sql.Timestamp or similar eventually. And *that* is much more tricky
that I, at least, originally thought.
Among other quirks:
- creating a Timestamp from a year/day/etc. assumes the given
year/day/etc. are in the default JVM timezone.
- the JDBC driver will sometimes extract the year/day/etc. and
interpret them as being in the DB timezone, or will sometimes use a
DateFormat with a timezone to convert it to the correct timezone. It
depends on the driver and even the version of the driver.
- java.sql.Timestamp and java.time do not rely on the same calendar:
Julian/Gregorian calendar for one, proleptic Gregorian calendar for the
other.
- java.sql.Timestamp and java.time do not assume the same offsets for
various zone IDs around and before 1900, when time zones were not a
formalized concept.
I've spent days on conversion problems between java.time and java.sql in
ORM over the last few months.
Which is why I think using LocalDateTimeType to convert between the
LocalDateTime literal and the Timestamp would be a good idea. If you want
to rewrite that code for literals, sure that can work, but exhaustive
testing will be needed.
> As counter-intuitive as it sounds, a ZonedDateTime actually includes an
offset to differentiate the overlap case you mention.
Yep. That's why it accepts parsing a ZoneDateTime with both a zone ID and
an offset. Try this:
https://gist.github.com/yrodiere/278996f865a9854e222aea58b5a7f26e
Note that a bug affects parsing ZoneDateTimes with both offset and zone ID
on JDK8 (fixed in 9):
https://bugs.openjdk.java.net/browse/JDK-8066982
We have a helper to work around that in Search:
https://github.com/hibernate/hibernate-search/blob/334e4aad5c776151bcf5db...
> I think the confusion here is in terms of (1) recognizing a temporal
literal and "parsing it" and (2) applying it to SQL. Different parts of
the puzzle.
Yep.
Yoann Rodière
Hibernate Team
yoann(a)hibernate.org
On Tue, 7 Jan 2020 at 19:50, Steve Ebersole <steve(a)hibernate.org> wrote:
> As far as I know, even Java does not support that. A true zone-id would
>> be something like (for me) "America/Chicago". If I ask Java to parse
>> "2020-01-01 10:10:10 America/Chicago +02:00" it just says no. For me,
CST
>> (standard) and CDT (daylight savings) are really synonyms for offset -
>> either UTC-05:00 or UTC-06:00 depending on day of the year.
>>
>
> It seems like the proper syntax for that would actually be "2020-01-01
> 10:10:10+02:00 America/Chicago", but in my
> testing DateTimeFormatter#parseBest did not handle that form either
>
>