On Sat 18 Mar 2017 at 21:58, Christian Schulte <c...@schulte.it> wrote:
> Am 03/18/17 um 14:04 schrieb Stephen Connolly:
> > Given that these new methods actually have implementations, could we see
> > about having at least a unit test of the new code - since it will not be
> > covered by any test.
>
> There would be an integration test for the code. Since nothing uses it,
> no way to add anything to the ITs.
>
> >
> > If we have no test and no usage, then we could realistically replace the
> > implementations with `throw new UnsupportedOperationsException("Not
> > implemented yet");` which would defeat the claimed purpose on the JIRA of
> > making this code available to IDE integrators.
>
> Throwing an UnsupportedOperationException would be an option. Reverting
> the commit also would be an option.
>
> > So I think to make "live" code available, we need at least a sanity check
> > of the new code with a unit test or two... nothing fancy.
>
> Just tell me what I should do.
>
> a) revert the commit and postpone to 3.6.0.
> b) remove the implementation and throw an UnsupportedOperationExceptio
Can you write a unit test (perhaps with mocks) showing that it does
effectively the same as the variant taking Project)?
That'd be good enough for 3.5.0
>
> I do not know how the public Maven API has been versioned in the past.
> The method will be needed to add support for dependency management
> *import* version ranges to the model builder. If we would add this in
> 3.5.1, we would add methods to an interface of Maven's public API in a
> patch release. No one would expect this to happen. We have done this in
> the past. Just take a look at the history of the 'ModelResolver'
> interface. This is something we should try to avoid to happen as much as
> possible.
We do not claim to be following semver.
Semver is not great for a large multi-component system.
If you have a small library, semver makes sense... there is a single
component and people are just consumers of the API so breaking changes are
important.
Now Maven is a build tool chain... we have different classes of consumer:
* users who interact via the pom
* plugin authors who interact via the plugin api
* ide integrators who interact via a different api
* 3rd parties consuming some of core as libraries
* etc
Changes will be "breaking" for some consumers and not for others.
We cannot be "semver" for all those consumers otherwise we would be on
Maven 375.0.0
If we are even close to semver it will. E for the biggest group of
consumers: i.e. Our users.
If we change the dependency resolution in a backwards incompatible way such
that their existing pom will not build... they want to know about it.
So where we are semver-like is the version number we give *to the users*
Everyone else are considered to be a smaller set of people that have to
just suck it up and read the release notes.
So if we are adding a new method to help ide integrators in 3.5.1... whoop
de doo. The user's don't care about that enough to bump it to 4.0.0 (yep
adding a method to an interface pre Java 8 is a *breaking change* if strict
semver we'd have to go to 4.0.0... I told you we weren't strict semver)
>
> Regards,
> --
> Christian
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscr...@maven.apache.org
> For additional commands, e-mail: dev-h...@maven.apache.org
>
> --
Sent from my phone
