Index is never set for Collection and Array in InvalidFormatException.Reference

[fdaugan]

When a InvalidFormatException is created, index values is always '-1'.
Indeed, in StringCollectionDeserializer, and CollectionDeserializer the exception is not caught.
The JsonMappingException shoud be caught and the index should be added and based on the "result" size.
Without this information, there is no way to get the index of the item involved in the mapping error.

[cowtowncoder]

Thank you for reporting this. Intent is indeed to report correct index...

[cowtowncoder]

Forgot one thing: this has only been supported on serialization side, since that is where it is more obviously needed.

But it would be nice on deserialization too, given that generally collection entry counts and JSON index (which would be available via JsonParser) align, and since actual JsonParser instance may or may not be available. Also, in case of nested JSON Arrays, locating index of problematic container might be tricky.

[cowtowncoder]

Fixed for 4 relevant deserializers; please let me know if I missed something. Also added basic unit tests.

[fdaugan]

I've just tested this change and this is perfect. You have added the missing reference of collection/array and attached the index to it.
It's perfect, now we have the entire path.

Thanks.

[fdaugan]

Just on thing, the index is well set, but the "InvalidFormatException.Reference#_from" is a Class instead of the true Collection/Array reference.
Having this information could be usefull to see the previous successfully parsed collection's items.

[cowtowncoder]

@fabdouglas Right, the problem with array is that one is only instantiated after collecting all entries (since arrays are not resizable). With collections this may not be the case, so I should go back and add a reference if possible. I realize that non-type-safety between Class and instance is awkward, but sometimes that is best that can be done when deserializing (f.ex. if problem occurs during deserialization of creator parameter, or due to buffering).

[cowtowncoder]

One possible thing to do would be to create incomplete arrays, although this would only be for purposes of error reporting. But I am wondering if this might not actually be a good thing: it'd be possible to indicate (for example) what was the preceding non-erroneous element.
So perhaps it would be possible to further improve this.

[fdaugan]

Indeed, the array copy may cause serious performance issues for large objects.
The main goal of this issue was the path making possible for JSON client to spot the incorrect data

[fdaugan]

BTW, the target Milestone is currently 1.9.13 instead of 2.4.2

[cowtowncoder]

@fabdouglas Thanks, fixed milestone. And yes, perhaps consider further improvements later on; while constructing array can be useful, it could also be problematic, so might want to use heuristics to determine that... i.e. needs more thinking.

[tea-dragon]

There are a couple places you use ix as the index that I think might be better represented as ix + buffer.bufferedSize().