I think I see another reason to allow type inferencing, which is less user code starts failing to compile when adopting 2.8.0 from 2.7.0. This is because user code in 2.7.0 could have done:
FSIterator<MyAnnot> = myJCas.getJfsRepository().getAllIndexedFS(MyAnnot.type); and it would work in 2.7.0 (because getAllIndexedFS(type) returned FSIterator (no generic info), which Java can assign to FSIterator<MyAnnot> with a warning. But without type inference, and returning FSIterator<TOP>, I think this would now be a hard failure, and user code that used to work would now start not compiling. -Marshall On 7/8/2015 3:47 PM, Marshall Schor wrote: > OK. Based on the previous discussion in this thread, and at least one other > person posting their view in favor of type inferencing (vs. safety), I'm going > to allow (3) to have type inference. > > -Marshall > > On 7/8/2015 11:17 AM, Richard Eckart de Castilho wrote: >> On 08.07.2015, at 17:11, Marshall Schor <m...@schor.com> wrote: >> >>> I think that similar arguments apply to other getters, such as >>> getAnnotationIndex(some-type-spec). >>> >>> These are currently defined to be generic methods so they do type >>> inferencing. >> True. >> >> Mind, I'm arguing for the conservative and safe choice here ( typically not >> my role ;) ). >> >> Back in the issue [1] I had first chosen a signature that allowed non-safe >> type inference >> and you were cautious of that. >> >> But my feeling is that for the present case being cautious/conservative and >> in particular safe is the better choice. >> >> Cheers, >> >> -- Richard >> >> [1] >> https://issues.apache.org/jira/browse/UIMA-4299?focusedCommentId=14487976&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14487976 >
