signatures that are recorded for default methods (original) (raw)

Chris Hegarty chris.hegarty at oracle.com
Fri Dec 14 09:54:01 UTC 2012

On 14/12/2012 09:41, Peter Levart wrote:

On 12/14/2012 10:06 AM, Peter Levart wrote:

On 12/14/2012 07:21 AM, David Holmes wrote:

Paul,

On 14/12/2012 9:46 AM, Paul Benedict wrote: Lance,

Good questions. Someone with authority will surely answer, but here's my armchair opinion... If the Javadoc is to specify how the default method executes, than that would naturally infer all default implementations must have a stated contract. You can't document the default execution behavior in the public API and then let a provider switch the behavior. The two go hand-in-hand, imo. I couldn't really make sense of that. :) The method has a contract. The "default implementation" has to honor that contract. The question is whether how the "default implementation" honors the method contract is required to be part of a second contract. I hope that made sense :) I think that the answer is obvious. A default method provider has only as much freedom in choosing the implementation of the default method that particular implementation differences among various providers are not observable by the "sane" usage of the API. The only soft aspect might be performance. Any other behavioural difference should not be allowed. Otherwise there will be in-compatibilities among platform providers. For example, the default Iterator.remove() is implemented in Oracle's JDK as always throwing UnsupportedOperationException. The TCK should test for that and the Javadoc should specify that. Ok, I admit that in this particular case and other similar cases (where the default method does nothing useful), the specification could alternatively specify: "The default method behaviour is unspecified and useless. Don't call that method or make sure it is always overridden if you call it" - the TCK in that case doesn't test the behaviour of such

And forever more ever concrete Iterator implementation, that does not support remove, will have to implement the remove method to throw UOE. I think this is a big loss.

-Chris.

method.

Peter

As Joe said, default interface methods are no different than any other overridable methods. They have a contract and behaviour. The behaviour can be changed (overriden) within constraints defined by contract, but the behaviour itself should also be specified and followed by different providers. Just my 2 cents. Regards, Peter

David -----

Paul On Thu, Dec 13, 2012 at 5:30 PM, Lance Andersen - Oracle <Lance.Andersen at oracle.com> wrote: On Dec 13, 2012, at 6:16 PM, Leonid Arbuzov wrote:

Good point, Joe. Those extra assertions for default methods can be checked by regular API tests separately from signature tests. I am not clear on this. See the message attached from David which ask a similar question (from the attached email): ------------------- 2. It is not obvious to me that the JDK's choice for a default implementation has to be the only possible implementation choice. In many/most cases there will be a very obvious choice, but that doesn't mean that all suppliers of OpenJDK classes have to be locked in to that choice. -------------------

If everyone needs to implement the same default implementation then great the JCK/TCK can test it and IMHO we should have a javadoc tag for this. If everyone is free to choose what the default behavior is, then we cannot test it. So has there been a decision WRT the requirements for default methods? Best Lance Thanks, -leonid On 12/13/2012 1:05 PM, Joe Darcy wrote: Hello, As with concrete methods on abstract classes, I would expect the specifications of the default methods to often contain text akin to "This default implementation does x, y, and z" since if the method is to be called by subtypes, the self-use patterns in the default method need to be known. Cheers, -Joe On 12/13/2012 11:24 AM, Leonid Arbouzov wrote: Hello Lance,

My understanding would be that the signature test should check that interface method is marked as default method but do not track the code in its default body (assuming that the body is not a part of a spec - API javadoc). (I've confirmed that with the signature test developer) Thanks, -leonid On 12/6/2012 9:01 AM, Lance Andersen - Oracle wrote: Folks,

Will the signatures for interfaces that are recorded by the TCKs for interfaces record the fact that a method includes a default method? or will it just record the method definition? I am assuming it will, but I know there has been discussion that a implementor could choose a different default implementation on one of the recent threads that was up for review. I am still trying to understand what our guidelines are, if any for documenting the behavior of the supplied default methods given the javadocs are part of the specification for many APIs (and in some case the only spec). Best Lance Lance Andersen| Principal Member of Technical Staff | +1.781.442.2037 Oracle Java Engineering 1 Network Drive Burlington, MA 01803 Lance.Andersen at oracle.com

Lance Andersen| Principal Member of Technical Staff | +1.781.442.2037 Oracle Java Engineering 1 Network Drive Burlington, MA 01803 Lance.Andersen at oracle.com

More information about the core-libs-dev mailing list