Hi, Thanks for the update. I think it makes sense to expand testing slightly, testing all three parsing clauses that you fixed, and for all three of them also checking that a "thing" after the broken item is parsed correctly.
But, the reason we didn't fix this the last two times we looked at it (that I am aware of) is the compatibility story. In order to fix his you need to figure out two things: - Is this is a spec change, that is, can we get away with throwing an AnnotationFormatError where we would now do? Should we clarify the spec? - Since this is a behaviorally incompatible change, how big is the impact? This is of course a hard question to answer, but if one could do a corpus analysis over a large code base and look for catches of ArrayStoreExceptions when reflecting over annotations, that could be useful. If it turns out that "a lot" of users have adopted to this bug, perhaps it isn't worth fixing? On the other hand I can imagine that this is so uncommon that no one catches either type of error. I can't help you with the internal processes at Oracle anymore, but if you are interested in doing a compatibility analysis I can make sure it reaches the right people. cheers /Joel On Thu, Jan 7, 2016 at 7:59 PM, Liam Miller-Cushon <cus...@google.com> wrote: > Hi Joel - > > On Tue, Jan 5, 2016 at 1:16 PM, Joel Borggrén-Franck > <joel.borggren.fra...@gmail.com> wrote: >> >> I think you need to do skipMemberValue the correct number of times so >> that the cursor in buf is correct, or? > > > Thanks for the catch. I added your test case, and the exception proxy is now > stored and returned after the entire array is read. > >> >> Also, in MissingArrayElementTest, why not just rethrow t? > > > Done. > > Thanks!