This is complex to solve for the client (and kind of unexpected that this is not working). So what speaks against supporting this use case properly in Sling. Maybe you can elaborate a bit on > I think calling resolve() for traversal is totally unexpected.
IMHO the complex cases should be nicely covered by Sling (instead of adding that burden to each client). Of course calling resolve() would only be done for the NonExistingResource (and maybe the SyntheticResource). The handling for regular existing resources should stay exactly as is (i.e. not relying on resolve() at all). I meanwhile opened https://issues.apache.org/jira/browse/SLING-5773 <https://issues.apache.org/jira/browse/SLING-5773> about this problem. Konrad > On 10 Jun 2016, at 09:26, Carsten Ziegeler <cziege...@apache.org> wrote: > > I think calling resolve() for traversal is totally unexpected. Now > looking at this, I'm really wondering if it was a good idea to implement > getParent on NonExistingResource. Let the client deal with it instead of > adding complex special cases. > > Carsten > > Konrad Windszus wrote >> There is one related problem with mapping: >> >> Consider the case you have a resource resolver mapping from "/content/" to >> "/". In the underlying repository you have a resource in >> "/content/existingParent". >> >> Now it might happen that a NonExistingResource is resolved for path >> "/existingParent/nonExistingResource". Since this is pointing towards a >> non-existing resource the mapping is not active (i.e. the path will not >> contain "/content"). So you end up with a NonExistingResource with path >> "/existingParent/nonExistingResource". >> Now you call getParent() on this resource. >> >> You would expect to end up with an existing resource in >> "/content/existingParent" (because that path does exist in the repository). >> Unfortunately due to the logic in AbstractResource.getParent() this just >> calls ResourceResolver.getParent() which will not use >> ResourceResolver.resolve(...) but rather ResourceResolver.getResource(...) >> internally. So the mapping is not considered here. Therefore no resource is >> found at "/existingParent", although it does exist at >> "/content/existingParent". >> >> This is unexpected and wrong in my opinion. So the NonExistingResource >> should not rely on AbstractResource.getParent() at all, but rather use its >> own implementation relying on ResourceResolver.resolve(...). WDYT? >> >> Thanks, >> Konrad >> >> >>> On 02 Jun 2016, at 14:48, Konrad Windszus <konra...@gmx.de> wrote: >>> >>> Thanks for the input. >>> I created https://issues.apache.org/jira/browse/SLING-5757 for tracking >>> that and I am going to propose a patch in there. >>> >>> What about SyntheticResources which are not NonExistingResources? If we >>> would follow the same approach as for NonExistingResources the question is, >>> with which resource type the non-existing parent resource should be >>> created? Or should a SyntheticResource (which is not a NonExistingResource) >>> return a NonExistingResource for getParent() instead in that case? >>> >>> Konrad >>> >>> >>>> On 02 Jun 2016, at 13:47, Daniel Klco <dk...@apache.org> wrote: >>>> >>>> I would imagine that the only thing this would change is to make a small >>>> number of null checks irrelevant. >>>> >>>> +1 for making the behavior more consistent, however, the JavaDocs and >>>> release notes should be explicit about this change. >>>> >>>> On Thu, Jun 2, 2016 at 4:45 AM, Georg Henzler <slin...@ghenzler.de> wrote: >>>> >>>>> Hi Konrad, >>>>> >>>>> +1 for making the behaviour of NonExistingResource more consistent - I >>>>> personally can't think of any places this would break existing code. >>>>> >>>>> Regards >>>>> Georg >>>>> >>>>> >>>>> >>>>> On 2016-06-01 15:09, Konrad Windszus wrote: >>>>> >>>>>> Hi Robert, >>>>>> thanks for your input. >>>>>> >>>>>> >>>>>>> I am not sure whether this would confuse existing clients though... >>>>>>> >>>>>> >>>>>> I am also a bit worried about that but the only example I could think >>>>>> of is a code trying to create the parent nodes or collecting the >>>>>> non-existing ones by checking getParent() for null. >>>>>> >>>>>> This would be pretty bad style IMHO therefore I would deliberately be >>>>>> willing to break that code. I wonder what do others think about >>>>>> changing the semantics of getParent() for NonExistingResource and >>>>>> probably also SyntheticResource. >>>>>> Konrad >>>>>> >>>>> >>>>> >>> >> >> > > > > -- > Carsten Ziegeler > Adobe Research Switzerland > cziege...@apache.org
