The most obvious remaining question appears to be: how do we spell
"permits <nest-members>".
We could say "permits this", which has the advantage of being a keyword,
but doesn't really mean what it says.
We could say "permits local", though "local" usually means local to a
method.
We could say "permits nest" or "permits nested", though if we decide to
allow public auxilliary subclasses in this case, then they are not
likely to be nestmates, and surely not nested.
We could allow the permits clause to be omitted, and be interpreted as
"permits nest", but this would much more severely change the meaning of
final -- people have an idea what final means, and this surely wouldn't
be it.
Open to other ideas...
(This is one of the costs of retconning final; if we used "sealed", we
wouldn't need a permits clause at all. Overall its probably still a
good tradeoff, but we do have to paint this shed acceptably.)
On 12/7/2018 11:38 AM, Brian Goetz wrote:
I’ve updated the document on sealing to reflect the discussion so far.
Sealed Classes
*Definition.* A /sealed type/ is one for which subclassing is
restricted according to guidance specified with the type’s
declaration; finality can be considered a degenerate form of sealing,
where no subclasses at all are permitted. Sealed types are a sensible
means of modeling /algebraic sum types/ in a nominal type hierarchy;
they go nicely with records (/algebraic product types/), though are
also useful on their own.
Sealing serves two distinct purposes. The first, and more obvious, is
that it restricts who can be a subtype. This is largely a
declaration-site concern, where an API owner wants to defend the
integrity of their API. The other is that it potentially enables
exhaustiveness analysis at the use site when switching over sealed
types (and possibly other features.) This is less obvious, and the
benefit is contingent on some other things, but is valuable as it
enables better compile-time type checking.
*Declaration.* We specify that a class is sealed by applying the
|final| modifier to a class, abstract class, interface, or record, and
specifying a |permits| list:
|final interface Node permits A, B, C { ... } |
In this explicit form, |Node| may be extended only by the types
enumerated in the |permits| list (which must further be members of the
same package or module.)
In many situations, this may be overly explicit; if all the subtypes
are declared in the same compilation unit, we may wish to permit a
streamlined form of the |permits| clause, that means “may be extended
by classes in the same compilation unit.”
|final interface Node permits __nestmates { ... } |
(As usual, pseudo-keywords beginning with |__| are placeholders to
illustrate the overall shape of the syntax.)
We can think of the simpler form as merely inferring the full
|permits| clause from information already present in the source file.
Anonymous subclasses (and lambdas) of a sealed type are prohibited.
*Exhaustiveness.* One of the benefits of sealing is that the compiler
can enumerate the permitted subtypes of a sealed type; this in turn
lets us perform exhaustiveness analysis when switching over patterns
involving sealed types. (In the simplified form, the compiler computes
the permits list by enumerating the subtypes in the nest when the nest
is declared, since they are in a single compilation unit.)
/Note:/ It is superficially tempting to say |permits package| or
|permits module| as a shorthand, which would allow for a type to be
extended by package-mates or module-mates without listing them all.
However, this would undermine the compiler’s ability to reason about
exhaustiveness, because packages and modules are not always
co-compiled. This would achieve the desired subclassing restrictions,
but not the desired ability to reason about exhaustiveness.
*Classfile.* In the classfile, a sealed type is identified with an
|ACC_FINAL| accessibility bit, and a |PermittedSubtypes| attribute
which contains a list of permitted subtypes (similar in structure to
the nestmate attributes.) Classes with |ACC_FINAL| but without
|PermittedSubtypes| behave like traditional final classes.
*Sealing is inherited.* Unless otherwise specified, abstract subtypes
of sealed types are implicitly sealed, and concrete subtypes are
implicitly final. This can be reversed by explicitly modifying the
subtype with |non-final|.
Unsealing a subtype in a hierarchy doesn’t undermine all the benefits
of sealing, because the (possibly inferred) set of explicitly
permitted subtypes still constitutes a total covering. However, users
who know about unsealed subtypes can use this information to their
benefit (much like we do with exceptions today; you can catch
FileNotFoundException separately from IOException if you want, but
don’t have to.)
/Note:/ Scala made the opposite choice with respect to inheritance,
requiring sealing to be opted into at all levels. This is widely
believed to be a source of bugs; it is relatively rare that one
actually wants a subtype of a sealed type to not be sealed, and in
those cases, is best to be explicit. Not inheriting would be a simpler
rule, but I’d rather not add to the list of “things for which Java got
the defaults wrong.”
An example of where explicit unsealing (and private subtypes) is
useful can be found in the JEP-334 API:
|final interface ConstantDesc permits String, Integer, Float, Long,
Double, ClassDesc, MethodTypeDesc, MethodHandleDesc,
DynamicConstantDesc { } final interface ClassDesc extends ConstantDesc
permits PrimitiveClassDescImpl, ReferenceClassDescImpl { } private
class PrimitiveClassDescImpl implements ClassDesc { } private class
ReferenceClassDescImpl implements ClassDesc { } final interface
MethodTypeDesc extends ConstantDesc permits MethodTypeDescImpl { }
final interface MethodHandleDesc extends ConstantDesc permits
DirectMethodHandleDesc, MethodHandleDescImpl { } final interface
DirectMethodHandleDesc extends MethodHandleDesc permits
DirectMethodHandleDescImpl { } // designed for subclassing non-final
class DynamicConstantDesc extends ConstantDesc { ... } |
*Enforcement.* Both the compiler and JVM should enforce sealing, as
they both enforce finality today (though from a project-management
standpoint, it might be allowable for VM support to follow in a later
version, rather than delaying the feature entirely.)
*Accessibility.* Subtypes need not be as accessible as the sealed
parent. In this case, some clients are not going to get the chance to
exhaustively switch over them; they’ll have to make these switches
exhaustive with a |default| clause or other total pattern. When
compiling a switch over such a sealed type, the compiler can provide a
useful error message (“I know this is a sealed type, but I can’t
provide full exhaustiveness checking here because you can’t see all
the subtypes, so you still need a default.”)
*Javadoc.* The list of permitted subtypes should be incorporated into
the Javadoc. Note that this is not exactly the same as the current
“All implementing classes” list that Javadoc currently includes, so a
list like “All permitted subtypes” might be added (possibly with some
indication if the subtype is less accessible than the parent, or
including an annotation that there exist others that are not listed.)
/Open question:/ With the advent of records, which allow us to define
classes in a single line, the “one class per file” rule starts to seem
both a little silly, and constrain the user’s ability to put related
definitions together (which may be more readable) while exporting a
flat namespace in the public API. I think it is worth considering
relaxing this rule to permit for sealed classes, say: allowing public
auxilliary subtypes of the primary type, if the primary type is public
and sealed.
