[
https://issues.apache.org/jira/browse/GROOVY-12319?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18109906#comment-18109906
]
ASF GitHub Bot commented on GROOVY-12319:
-----------------------------------------
daniellansun commented on PR #2845:
URL: https://github.com/apache/groovy/pull/2845#issuecomment-5479443071
@paulk-asert Thanks, Paul — this was the right list to walk through before
merge. We have decided each row as below; the well-formedness checks stay on
**every** compilation unit (dynamic Groovy and scripts included), with no
opt-out.
### Team call 1 — `new T[n]` / `new T[]{…}`
We will **reject** it, with no carve-out in `isReifiable`.
A type variable is not reifiable (JLS 4.7 / 15.10.1). What Groovy 4/5
compiled was `Object[]` (or the bound) typed as `T[]` — Java’s `(T[]) new
Object[n]`, without the cast in the source. Treating `T` as reifiable would
also be the wrong predicate for `instanceof T`.
The Java idiom remains the workaround, in both dynamic and `@CompileStatic`
code:
```groovy
items = (T[]) new Object[capacity] // unbounded T
items = (T[]) new Bound[capacity] // T extends Bound
items = (T[]) Array.newInstance(token, n) // when a Class<T> is at hand
```
`new List<String>[n]` stays rejected; `new List<?>[n]` stays legal.
### Team call 2 — generic class extending `Throwable`
We will **reject** it as well, for JLS 8.1.2 / javac parity. Groovy’s
`catch` is erased, so the class would still run on the JVM — but this PR’s
charter is well-formedness that javac already enforces, and Groovy 6 is the
window.
Workaround: keep the exception non-generic and move the type parameter onto
an accessor (or a factory). Parameterized `catch` is already impossible in the
grammar (`catchType` has no type arguments).
### Remaining rows
Agreed: **break and document** for all of them, including `List<int>`, `new
T()` / `T.class`, `implements X<? extends …>`, `new ArrayList<?>()` , a class
type as an additional bound, `Foo<String>.class`, and `Outer.Inner<String>`
with a raw generic `Outer`.
A table in that shape is now in `core-object-orientation.adoc` §Generics
(`[[generics-well-formedness]]`), with the two workarounds above as executable
examples. I would still welcome a matching table in the 6.0 release notes if
you would like to take that, as you offered.
### Message nits
Both fixed:
- `new T[n]` now reports `generic array creation of T` (the type variable,
not the erasure).
- `Foo<String>.class` now reports `Cannot select from a parameterized type`
(JLS 15.8.2). `Cell<String>.ID` / `.id()` keep the 4.5.2 “static member through
a parameterization” wording. `value.class` / `value*.class` are unchanged —
those are `getClass()`, not class literals.
### `"outer.class"` metadata
Noted. The enclosing type lives on the `ClassNode` field
(`getOuterClassType()` / `@since 6.0.0`); we will not restore a metadata
fallback. Transform authors who still read the GROOVY-10646-era key will see
`null`.
> Java compatibility: remaining generic type syntax
> -------------------------------------------------
>
> Key: GROOVY-12319
> URL: https://issues.apache.org/jira/browse/GROOVY-12319
> Project: Groovy
> Issue Type: Improvement
> Reporter: Daniel Sun
> Priority: Major
>
> Groovy already accepts most Java generics. Three Java forms still fail: two
> in the parser, one in {{GenericsVisitor}}. Each has a local workaround.
> Method type arguments such as {{Helper.<String>identity( x )}} already work
> and are not part of this request.
> h3. 1. Diamond {{<>}} on an anonymous class
> Java 9+ (JEP 213, JLS 15.9.5) allows diamond when creating an anonymous
> class, if the target type supplies the type arguments:
> {code:java}
> Processor<String> p = new Processor<>() {
> public String process(String val) { return val.toUpperCase(); }
> };
> {code}
> This should work in assignment, as a method argument, and under
> {{@TypeChecked}} / {{@CompileStatic}}, for both interfaces and abstract
> classes.
> *Actual:* Groovy rejects it with {{Cannot use diamond <> with anonymous inner
> classes}}.
> *Workaround:* write the type arguments explicitly:
> {code:java}
> Processor<String> p = new Processor<String>() {
> public String process(String val) { return val.toUpperCase(); }
> };
> {code}
> GROOVY-6730 and GROOVY-7159 were false-positive STC errors when diamond was
> _not_ used. They did not add this form.
> h3. 2. Qualified parameterized inner types ("rare" types)
> Java allows an inner type to keep the enclosing type's arguments (JLS 4.5):
> {code:java}
> class Outer<T> {
> class Inner<U> {}
> }
> Outer<String>.Inner<Integer> x = new Outer<String>().new Inner<Integer>("v",
> 1);
> {code}
> The same qualification appears as a field or method type, nested in another
> type argument, as a superclass when the member is a non-static class, and in
> a qualified instance creation inside the outer class:
> {code:java}
> List<Outer<String>.Inner<Integer>> list;
> class Sub extends Outer<String>.Inner {
> Sub(Outer<String> o) { o.super(); }
> }
> // inside Outer:
> new Outer<T>.Inner<U>(...)
> {code}
> Selecting a static member type from a parameterization is a compile-time
> error (JLS 4.5.2 / 6.5.5). Nested interfaces and enums are implicitly static
> (JLS 9.5), so a nested interface must be selected from the raw enclosing name:
> {code:java}
> class Impl implements Outer.Inner<Integer> { ... } // legal
> class Bad implements Outer<String>.Inner<Integer> { ... } // error
> {code}
> *Actual:* Groovy fails to parse {{Outer<String>.Inner}} ({{Unexpected
> input}}).
> *Workaround:* a factory that returns {{Inner}} without naming
> {{Outer<T>.Inner}}.
> h3. 3. Explicit type arguments on constructors, {{this()}} and {{super()}}
> Java allows constructor type arguments independently of the class type
> arguments (JLS 15.9 / 8.8.7.1):
> {code:java}
> class Box {
> <T> Box(T t) {}
> Box() { <String>this("x"); }
> }
> class Derived extends Box {
> Derived() { <String>super("y"); }
> }
> class Outer {
> class Inner {
> <T> Inner(T t) {}
> }
> }
> new <String>Box("x");
> new Outer().new <String>Inner("z");
> {code}
> *Actual:* Groovy fails at {{new <}} with {{Unexpected input: '<'}}.
> *Workaround:* inference, e.g. {{new Box("x")}}.
> Constructor type arguments are already tracked by GROOVY-10501. The
> {{this()}} / {{super()}} / inner-{{new}} forms are the same JLS production
> and should be handled together.
> h3. Expected
> All three forms compile in dynamic Groovy and under {{@TypeChecked}} /
> {{@CompileStatic}}, matching javac on well-formed programs.
> The well-formedness rules are javac's (checked against javac 25). Parsing a
> rare type does not make every use of it legal.
> These remain legal. {{Inner}} is a non-static member of {{Outer}};
> {{Outer<?>}} is reifiable (JLS 4.7), so a non-static member type of that
> enclosing type is reifiable (JLS 15.10.1):
> {code:java}
> class Outer<T> {
> class Inner {}
> class InnerG<U> {}
> interface Iface<U> { U id(U u); }
> }
> Outer<?>.Inner[] a = new Outer<?>.Inner[0];
> Outer<?>.InnerG<?>[] b = new Outer<?>.InnerG<?>[0];
> Outer<?>.Inner field;
> Outer<String>.Inner concrete;
> class Impl implements Outer.Iface<Integer> {
> public Integer id(Integer u) { return u; }
> }
> {code}
> These remain compile errors:
> {code:java}
> new Object<>() {} // diamond on a non-generic type
> class C extends ArrayList<> {} // diamond on a class declaration
> List<> list; // diamond on a field
> new <String>Box<>("x") // diamond combined with
> constructor type arguments
> Outer<String, Integer>.Inner x; // wrong arity
> x instanceof Outer<String>.Inner // parameterized type is not
> reifiable
> new Outer<String>.Inner[1] // generic array creation (JLS
> 15.10.1)
> new Outer<?>.Nested[0] // static member from a
> parameterized type (JLS 6.5.5)
> Outer<?>.Nested z; // same rule as a type name
> new Outer<?>.InnerG[0] // raw generic member of a
> parameterized enclosing type
> class Bad implements Outer<String>.Iface<Integer> {} // nested interface is
> implicitly static (JLS 9.5)
> new java.util.Map<?,?>.Entry[0] // Map.Entry is a nested interface
> new Outer<?>().new Inner() // constructor type argument may
> not be a wildcard
> {code}
> The {{Nested}} cases assume {{static class Nested}} inside {{Outer}}. The
> {{InnerG}} array error is the raw-member form; the legal counterpart is {{new
> Outer<?>.InnerG<?>[0]}} above.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
