Just a note that this is still ready to be reviewed.
I was also told that JCK tests pass with the patch applied.
Regards, Peter
On 10/04/2016 03:40 PM, Peter Levart wrote:
Hi,
I have a fix for conformance (P2) bug (8062389
<https://bugs.openjdk.java.net/browse/JDK-8062389>) that also fixes a
conformance (P3) bug (8029459
<https://bugs.openjdk.java.net/browse/JDK-8029459>) and a performance
issue (8061950 <https://bugs.openjdk.java.net/browse/JDK-8061950>):
http://cr.openjdk.java.net/~plevart/jdk9-dev/Class.getMethods.new/webrev.04/
As Daniel Smith writes in 8029459
<https://bugs.openjdk.java.net/browse/JDK-8029459>, the following
situation is as expected:
interface I { void m(); }
interface J extends I { void m(); }
interface K extends J {}
K.class.getMethods() = [ J.m ]
But the following has a different result although it should probably
have the same:
interface I { void m(); }
interface J extends I { void m(); }
interface K extends I, J {}
K.class.getMethods() = [ I.m, J.m ]
He then suggests the following algorithm:
An implementation of getMethods consistent with JLS 8 would include
the following (see Lambda Spec, Part H, 9.4.1 and 8.4.8):
1) The class's/interface's declared (public) methods
2) The getMethods() of the superclass (if this is a class), minus any
that have a match in (1)
3) The getMethods() of each superinterface, minus any that have a
match in (1) or a _concrete_ match in (2) or a match from a
more-specific class/interface in (2) or (3)
But even that is not sufficient for the following situation:
interface E { void m(); }
interface F extends E { void m(); }
abstract class G implements E {}
abstract class H extends G implements F {}
H.class.getMethods() = [ E.m, F.m ]
The desired result of H.class.getMethods() = [ F.m ]
So a more precise definition is required which is implemented in the fix.
The getMethods() implementation partitions the union of the following
methods:
1) The class's/interface's declared public methods
2) The getMethods() of the superclass (if this is a class)
3) The non-static getMethods() of each direct superinterface
...into equivalence classes (sets) of methods with same signature
(return type, name, parameter types). Within each such set, only the
"most specific" methods are kept and others are discarded. The union
of the kept methods is the result.
The "more specific" is defined as a partial order within a set of
methods of same signature:
Method A is more specific than method B if:
- A is declared by a class and B is declared by an interface; or
- A is declared by the same type as or a subtype of B's declaring type
and both are either declared by classes or both by interfaces (clearly
if A and B are declared by the same type and have the same signature,
they are the same method)
If A and B are distinct elements from the set of methods with same
signature, then either:
- A is more specific than B; or
- B is more specific than A; or
- A and B are incomparable
A sub-set of "most specific" methods are the methods from the set
where for each such method M, there is no method N != M such that N is
"more specific" than M.
There can be more than one "most specific" method for a particular
signature as they can be inherited from multiple unrelated interfaces,
but:
- javac prevents compilation when among multiply inherited methods
with same signature there is at least one default method, so in
practice, getMethods() will only return multiple methods with same
signature if they are abstract interface methods. With one exception:
bridge methods that are created by javac for co-variant overrides are
default methods in interfaces. For example:
interface I { Object m(); }
interface J1 extends I { Map m(); }
interface J2 extends I { HashMap m(); }
interface K extends J1, J2 {}
K.class.getMethods() = [ default Object j1.m, abstract Map j1.m,
default Object j2.m, abstract HashMap j2.m ]
But this is an extreme case where one can still expect multiple
non-abstract methods with same signature, but different declaring
type, returned from getMethods().
In order to also fix 8062389
<https://bugs.openjdk.java.net/browse/JDK-8062389>, getMethod() and
getMethods() share the same consolidation logic that results in a set
of "most specific" methods. The partitioning in getMethods() is
partially performed by collecting Methods into a HashMap where the
keys are (name, parameter types) tuples and values are linked lists of
Method objects with possibly varying return and declaring types. The
consolidation, i.e. keeping only the set of most specific methods as
new methods are encountered, is performed only among methods in the
list that share same return type and also removes duplicates.
getMethod() uses only one such list, consolidates methods that match
given name and parameter types and returns the 1st method from the
resulting list that has the most specific return type.
That's it for algorithms used. As partitioning partially happens using
HashMap with (name, parameter types) keys, lists of methods that form
values are typically kept short with most of them containing only a
single method, so this fix also fixes performance issue 8061950
<https://bugs.openjdk.java.net/browse/JDK-8061950>.
The patch also contains some optimizations around redundant copying of
arrays and reflective objects.
The FilterNotMostSpecific jtreg test has been updated to accommodate
for changed behavior. Both of the above extreme cases have been added
to the test.
So, comments, please.
Regards, Peter
