Re: [VOTE] HttpComponents Core 4.2 release based on RC1

> On 24 April 2012 17:18, Oleg Kalnichevski <olegk@...> wrote:
> > On Tue, 2012-04-24 at 16:59 +0100, sebb wrote:
> >> On 24 April 2012 16:03, Oleg Kalnichevski <olegk@...> wrote:
> >> > On Tue, 2012-04-24 at 15:18 +0100, sebb wrote:
> >> >> On 24 April 2012 12:13, Oleg Kalnichevski <olegk@...> wrote:
> >> >> > On Tue, 2012-04-24 at 02:48 +0100, sebb wrote:
> >> >> >> On 23 April 2012 14:33, sebb <sebbaz@...> wrote:
> >> >> >> > On 21 April 2012 12:21, Oleg Kalnichevski <olegk@...> wrote:
> >> >> >> >> Please vote on releasing these packages as HttpComponents Core 4.2. The
> >> >> >> >> vote is open for the at least 72 hours, and only votes from
> >> >> >> >> HttpComponents PMC members are binding. The vote passes if at least
> >> >> >> >> three binding +1 votes are cast and there are more +1 than -1 votes.
> >> >> >> >>
> >> >> >> >> Packages:
> >> >> >> >> http://people.apache.org/~olegk/httpcore-4.2-RC1/
> >> >> >> >>
> >> >> >> >> Release notes:
> >> >> >> >> http://people.apache.org/~olegk/httpcore-4.2-RC1/RELEASE_NOTES.txt
> >> >> >> >>
> >> >> >> >> Maven artefacts:
> >> >> >> >> https://repository.apache.org/content/repositories/orgapachehttpcomponents-078/org/apache/httpcomponents/
> >> >> >> >>
> >> >> >> >> SVN tag:
> >> >> >> >> https://svn.apache.org/repos/asf/httpcomponents/httpcore/tags/4.2-RC1/
> >> >> >> >>
> >> >> >> >> --------------------------------------------------------------------------
> >> >> >> >>  Vote:  HttpComponents Core 4.2 release
> >> >> >> >>  [ ] +1 Release the packages as HttpComponents Core 4.2.
> >> >> >>
> >> >> >> Sorry, I'm changing my vote:
> >> >> >>
> >> >> >> >>  [X] -1 I am against releasing the packages (must include a reason).
> >> >> >>
> >> >> >> I've just noticed that Clirr reports several compatibility issues against 4.1.
> >> >> >>
> >> >> >> I've not investigated in any detail, but it looks as though at least
> >> >> >> some of these are binary compatibility issues, and they appear to be
> >> >> >> in public APIs.
> >> >> >>
> >> >> >> It may be that these are not actually a problem, but I think they need
> >> >> >> to be investigated further.
> >> >> >> If the errors are harmless - or perhaps only affect source builds - it
> >> >> >> would be helpful to update the site (and ideally release notes)
> >> >> >> accordingly.
> >> >> >>
> >> >> >> [No need to cancel the vote just yet, in case I'm wrong.]
> >> >> >>
> >> >> >> BTW, we recently added test jars to the Commons Maven output.
> >> >> >> This should make it easier to run old tests against new releases.
> >> >> >>
> >> >> >
> >> >> > Sebastian
> >> >> >
> >> >> > The reported differences in public APIs reported by Clirr are due to two
> >> >> > things (1) upgrade from Java 1.3 to Java 1.5 (2) removal of code
> >> >> > deprecated between 4.0-beta1 and 4.0 (that is, before 4.0 GA, more than
> >> >> > two years ago)
> >> >> >
> >> >> > We had a discussion about pros and cons of upgrading to Java 1.5 and if
> >> >> > I remember it correctly you were in favor of that idea [1]. The changes
> >> >> > have also been announced early enough (several releases in advance) [2].
> >> >> > They do make 4.1 and 4.2 not fully binary compatible but I seriously
> >> >> > doubt there will be a single user affected by incompatibility.
> >> >> >
> >> >> > I hope you will change your mind.
> >> >>
> >> >> I've been looking further at the changes.
> >> >> The changes to NIO are all removals of deprecated methods, so not a
> >> >> problem (or at least, not our problem).
> >> >>
> >> >> The removed methods in HttpCore are also deprecated methods, so not a problem.
> >> >>
> >> >
> >> > Not only were they deprecated, they are deprecated two release cycles
> >> > back (before 4.0 official release).
> >> >
> >> >> Not sure why the value definitions of HTTP.DEFAULT_CONTENT_CHARSET and
> >> >> DEFAULT_PROTOCOL_CHARSET were changed.
> >> >> Given that they are now deprecated, I would have thought the values
> >> >> could have been left untouched.
> >> >
> >> > I think the case changed (by mistake). I'll fix it right away.
> >> >
> >> >> However AFAICT that does not affect compatibility.
> >> >>
> >> >> [BTW, in future we ought to document in which release items are deprecated]
> >> >>
> >> >> That just leaves the changed method signatures, which are due to
> >> >> adding generics to Iterator in o.a.h.message.Basic*Iterator and to
> >> >> AbstractMessageParser.
> >> >> In the case of the MessageParser subclasses, these were also changed
> >> >> to use more specific subclasses:
> >> >> HttpRequest and HttpResponse instead of their common super-interface HttpMessage
> >> >>
> >> >> It's not obvious to me if these methods are likely to be called by 3rd
> >> >> party code or whether they are only likely to be used internally.
> >> >>
> >> >
> >> > You see, in any sane use case scenarios, especially as far as iterators
> >> > are concerned, the type returned from those methods would always be cast
> >> > to the expected subtype. In almost all cases regardless of how those
> >> > methods are being used the changes will have no effect at the runtime
> >> > behavior.
> >>
> >> The problem is that the return type of a method is part of the signature.
> >> Java won't find the method at runtime if the signature changes between
> >> compilation and run-time.
> >>
> >> This generally does not affect source compatibility, but it does
> >> affect binary compat.
> >>
> >> We had this exact problem in Commons IO
> >> We wanted to change a method return from void to something else;
> >> however testing against pre-existing binaries showed that this broke
> >> binary compat.
> >>
> >
> > All right. I'll revert those changes.
>
> We are already making the assumption breaking the API is OK for
> long-deprecated methods, i.e. that user applications have migrated
> away from the deprecated methods.
>
> So if the methods in question are not likely to be used by 3rd party
> applications - are they effectively internal ? - we could consider
> releasing with such breaks in compat, provided that such changes are
> clearly documented.
>