[records] Compile-time constants can be used in local classes static members but cannot be used in records

[Tagir Valeev]

чт, 30 янв. 2020 г., 0:46 Mateusz Romanowski <romanowski.mateusz at gmail.com>:

> Hi Alex et al.,
>
> How would we understand an up-level `final Object lock = new Object();`
> then?
>

Constant expressions are strictly defined by spec (15.28). This is not
compile time constant.

With best regards,
Tagir Valeev


> Should we perhaps resurrect C-like `static final` variables inside methods
> to allow static nested non-constexpr variables?
>
> /Mateusz
>
>
> W dniu środa, 29 stycznia 2020 Alex Buckley <alex.buckley at oracle.com>
> napisał(a):
> > Note that the up-level variable in this case -- `final int x = 2;` -- is
> a constant variable, but not a static constant variable (or a static member
> of anything). Still, the simple name `x` is in scope in the body of nested
> {class, record} X, and counts as a constant expression, so RecordTest
> should compile in 15 once when JDK-8236189 is undone/relaxed.
> >
> > Alex
> >
> > On 1/29/2020 6:44 AM, Brian Goetz wrote:
> >>
> >> Tagir is right, and this is yet another aspect of our technical debt
> with regard to nesting.  Nested entities of all stripes (static vs
> instance, class vs interface vs record vs enum) can access static members
> from lexically enclosing contexts.
> >>
> >>
> >>
> >> On 1/29/2020 2:03 AM, Tagir Valeev wrote:
> >>>
> >>> But:
> >>>
> >>> 1. Compile time constant is not the state.
> >>> 2. Static field of local class also cannot capture any state, even
> >>> technically (because there's no place to store the state). Yet, this
> code
> >>> is allowed. I observe asymmetry here.
> >>>
> >>> With best regards,
> >>> Tagir Valeev.
> >>>
> >>> ср, 29 янв. 2020 г., 10:39 Vicente Romero <vicente.romero at oracle.com>:
> >>>
> >>>> Hi Tagir,
> >>>>
> >>>> This behavior is in accordance with the spec: a local record is not a
> >>>> local class, local records are not allowed to capture any state from
> its
> >>>> enclosing type. We plan to revisit this in the future not only for
> >>>> record but as a general effort on nested-ness but for the time being
> we
> >>>> preferred to let local records immune to any capture rule that applies
> >>>> to local classes
> >>>>
> >>>> Vicente
> >>>>
> >>>> On 1/28/20 10:01 PM, Tagir Valeev wrote:
> >>>>>
> >>>>> Hello!
> >>>>>
> >>>>> This code is perfectly compilable (javac 14-ea+33-1439):
> >>>>>
> >>>>> public class RecordTest {
> >>>>>       public static void main(String[] args) {
> >>>>>           final int x = 2;
> >>>>>           class X {
> >>>>>               static final int y = x;
> >>>>>           }
> >>>>>       }
> >>>>> }
> >>>>>
> >>>>> This is not (simply changed class to record, leaving everything else
> the
> >>>>
> >>>> same):
> >>>>>
> >>>>> public class RecordTest {
> >>>>>       public static void main(String[] args) {
> >>>>>           final int x = 2;
> >>>>>           record X() {
> >>>>>               static final int y = x;
> >>>>>           }
> >>>>>       }
> >>>>> }
> >>>>>
> >>>>> Error:(5, 34) java: non-static variable x cannot be referenced from a
> >>>>> static context
> >>>>>
> >>>>> This looks counter-intuitive to me: why static field in one case can
> >>>>> refer to compile-time constant and in another case cannot?
> >>>>>
> >>>>> With best regards,
> >>>>> Tagir Valeev
> >>>>
> >>
> >
>