On 13/10/2016 18:30, Peter Levart wrote:
Hi Paul, Alan,
I incorporated Paul's suggestions into new webrev:
http://cr.openjdk.java.net/~plevart/jdk9-dev/Class.getMethods.new/webrev.05/
This iteration also contains a nearly-exhaustive test of
Class.getMethods(): PublicMethodsTest. It is actually a test
generator. Given a Case template, it generates all variants of methods
for each of the types in the case. Case1 contains 4 interface method
variants ^ 3 interfaces * 4 class method variants ^ 3 classes = 4^6 =
4096 different sub-cases of which only 1379 compile. The results of
those 1379 sub-cases are persisted in the Case1.results file. Running
the test compares the persisted results with actual result of
executing each sub-case. When running this test on plain JDK 9
(without patch), the test finds 218 sub-cases where results differ:
http://cr.openjdk.java.net/~plevart/jdk9-dev/Class.getMethods.new/PublicMethodsTest.jtr
Looking at those differences gives the impression of the effects of
the patch.
Overall I think this looks very good. I mostly focused on the
PublicMethods implementation to satisfy myself that it does selects the
most specific methods. In passing I wonder if "combine" or "merge" might
be better than "consolidate" for method name and terminology, a minor
point of course.
Given the behavior change then I think we'll need to capture it in a
release notes. I can't think of any libraries or frameworks that might
have see but something might come out of the woodwork and would be nice
to be able to point to a summary.
I assume that copyright headers will be added before this is pushed.
Also there is a @SuppressWarnings arguments that I don't recognize (IDE
specific)? It would good to trim back some of the really long times in
the test too (there are some >150 char lines that will be a pain to
review side-by-side when there are future changes).
-Alan
