Yeah, ObjAttribute will definitely preserve "type" as a "read-through" property. Otherwise ROP won't work for instance (DbEntity data is not passed to the client). The only thing we'll be preventing is mapping attributes with different Java types to the same column (e.g. item #3 on my list - different subclasses in an inheritance tree attempting to map the same column differently).
Andrus On Jul 22, 2012, at 5:39 PM, Aristedes Maniatis wrote: > Do we ever want it to be possible to create ObjEntities which don't map to a > DbEntity at all? Or even just attributes which don't (yet) have a column in > the database? Perhaps they are a future enhancement and we aren't ready to > map them to the database, but we'd like to start writing code around them. > > How about ObjEntities which are just an interface? > > I am not saying that these are very powerful use-cases, but that's all I can > think of just now. A nice improvement with your suggested change would be > better validation in the modeler. We'd be able to alert the user to an > Integer mapped against a bigint. It would be nice if the modeler still also > showed the Java type in the ObjEntity view, even if this was read only. > > Ari > > > On 22/07/12 8:20pm, Andrus Adamchik wrote: >> I am thinking of redefining one of the mapping assumptions that was in >> Cayenne since day one. In 3.2 I want to move attribute java type from >> ObjAttribute to DbAttribute. The goal of this change is to improve >> consistency of the runtime model. Current separation of Java and DB >> attribute types causes a whole class of bugs and a whole class of hacks in >> the framework. >> >> E.g.: >> >> 1. Unrecognized non-standard type mapping. This one is discussed at the >> moment on the user list [1]. I suspect it has nothing to do with "custom" >> types, but rather with non-JDBC default mapping of DB data to Java, >> regardless of the Java type. >> >> 2. Hacks to recognize non-standard type mapping. When creating a DataRow, >> Cayenne would try to guess which ObjEntities might use this DataRow, and >> populate DataRows with values corresponding to the ObjAttribute type >> definitions. This clearly breaks layer separation - lower layers have to >> know too much about the higher layers of the stack. Besides it doesn't >> always work anyways - see #3. >> >> 3. Extra mapping "flexibility" that doesn't really work. We had past Jiras >> when the same column is mapped to different Java types in 2 different >> subclasses, creating a mess in subclass-agnostic DataRows. >> >> This is not a full list of problems, but gives you some idea. I am hoping >> the suggested change would tie things up and leave no space for ambiguities. >> >> Comments? >> >> Andrus >> >> [1] http://markmail.org/message/6bs2suislyfp3apk >> > > -- > --------------------------> > Aristedes Maniatis > GPG fingerprint CBFB 84B4 738D 4E87 5E5C 5EFA EF6A 7D2E 3E49 102A >
