hi Lars I just reply the Q you raised in HBASE-19746. (smile)
In short, it is impossible to keep the BC for the scenario you asked. That is way I prefer to introduce a new annotation to break the BC of inheritance. On 2018/04/22 04:18:03, Lars Francke <lars.fran...@gmail.com> wrote: > Sorry to dig up this old thread. I've also just commented on < > https://issues.apache.org/jira/browse/HBASE-19746> > > The book says: > > MAJOR version when you make incompatible API changes, > > So do we really need to keep BC between major versions? How else would we > ever add new methods to interfaces without having default implementations? > > On Fri, Dec 15, 2017 at 2:26 PM, Anoop John <anoop.hb...@gmail.com> wrote: > > > Good point Chia. Interfaces like Region, TableDesc are not meant for > > own impl by the users or CP users. These are just read only > > interfaces. Would be best if we can have a way to say so. We > > should distinguish an exposed interface/class to say whether this is > > just read only or allowed to be extended. > > > > In another jira we have a discuss abt the own impl of Cell being added > > to Put or not.. We are facing some issues if user do so.. So if we > > have this new annotation and mark clearly that Cell is just read only > > that will be a better way than just saying in javadoc. :-) But then > > we have APIs which takes interfaces like Cell/TableDescriptor as > > param. So some way we should say how that can be done.. We do have > > Factory in all such places. Nice high level idea Chia.. We can > > clearly say things in each and every exposed class/interfaces. Will > > make our life easy . > > > > -Anoop- > > > > On Fri, Dec 15, 2017 at 10:22 AM, Ted Yu <yuzhih...@gmail.com> wrote: > > > We use @InterfaceStability.Stable in our codebase. > > > > > > If backward compatibility is kept, isn't the class / interface Stable ? > > > > > > Cheers > > > > > > On Thu, Dec 14, 2017 at 8:48 PM, Chia-Ping Tsai <chia7...@apache.org> > > wrote: > > > > > >> hi folks, > > >> > > >> We have added many Public "read-only" interface in 2.0 to save user from > > >> the dangerous operations. For example, TableDescriptor, > > >> ColumnFamilyDescriptor, Region, Store, etc. However, this change make > > us be > > >> hard to enhance these interface (ie. add the new APIs) because the BC > > will > > >> be broken for the user having the subclass of these interfaces. In > > contrast > > >> with Cell, Filter, and coprocessor, these new interfaces are NOT > > designed > > >> to be a extendable class for user but we have no explicit caveat. As I > > see > > >> it, it is necessary to introduce an new annotation to explicitly say > > "you > > >> CAN extend this Public class with the BC guarantee". User should not > > extend > > >> the Public classes which don't have the such annotation. > > >> > > >> Any suggestions? > > >> > > >> Cheers > > >> Chia-Ping > > >> > > >
