[hibernate-dev] 6.0 - HQL literals

[Yoann Rodiere]

Hi,

The syntax looks nice.
I suppose it's future-proof enough, though I can imagine us getting in
trouble if JDBC starts allowing parameterized or custom formats, which may
start with a digit or even (in edge cases) look like a date. That seems
unlikely, so I think it's an acceptable risk.

I'm not entirely sure allowing JDBC literals to be "passed-through" the HQL
will always be intuitive, even if it's already allowed for integers: under
some circumstances we map date-only or time-only types to timestamps, for
example. The database might cast a date-only value to a timestamp
automatically by setting hours/minutes/etc to zero, but I'm not sure that's
what *we* do when persisting, considering the various hacks we have around
timezones; '2019-01-01' might very well be converted to
'2018-12-31T23:00:00', for all I know. As a result, there might be a broad
range of Java types where these literals will be seen as "buggy".

The solution would be to support litterals for the *java* types and
converting them to the correct SQL type automatically depending on the
context, but that will be a hassle, if even possible at all.

I might be wrong, but only exhaustive testing of all literals with all
date/time types on all RDBMS will let us know for sure. Let's keep in mind
how many bugs have surfaced from time-related features in the past...


Yoann Rodière
Hibernate Team
yoann at hibernate.org


On Mon, 6 Jan 2020 at 15:57, Steve Ebersole <steve at hibernate.org> wrote:

> Initially when I started working on 6.0 I added support for defining
> temporal literals using JDBC's "escape syntax".  JDBC already defines a
> syntax for declaring temporal literals using this syntax -
>
>    - {d '2020-01-01'} for a Date
>    - {t '10:10:10'} for a Time
>    - {ts '2020-01-01 10:10:10'} for a Timestamp
>
> I had planned on using this syntax to define generalized support for adding
> new types of literals using the "prefix" which is the first identifier
> after the open brace.  I did not have any concrete plans for specific types
> of literals, although I was hoping this would fit with hibernate-spatial
> needs.
>
> Since temporal values are so common I added another simplified form:
>
>    - {2020-01-01} for a "Date"
>    - {10:10:10} for a "Time"
>    - {2020-01-01 10:10:10} for a "Timestamp"
>
> Notice first the absence of quotes for these.  The patterns is defined as a
> syntactic element of the grammar (thanks to Gavin King for this particular
> idea)[1].
>
> As a side note, I first tried to use back-ticks for these simplified
> temporal literals rather than braces but that conflicts with the
> `QUOTED_IDENTIFIER` lexer rule.  `QUOTED_IDENTIFIER` is never used (and
> really kind of meaningless in the HQL grammar) so one option would be to
> remove that rule and use the back-ticks for these literals like
> `2020-01-01` as opposed to {2020-01-01} if folks like that better.
>
> It's also important to note that I actually use the Java 8 temporal types
> internally to represent these simplified literals because it is easy to
> translate from one type to another starting from these types.
>
> Anyone have objections or suggestions regarding any of this?
>
>
> [1] -
>
> https://github.com/hibernate/hibernate-orm/blob/eab6107ec2e7b3a0c06146a9ff51b9964f4b3169/hibernate-core/src/main/antlr/org/hibernate/grammars/hql/HqlParser.g4#L484
> _______________________________________________
> hibernate-dev mailing list
> hibernate-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/hibernate-dev
>
>