Exception while unmarshalling Integer with an element bigger than 32-bit

[Brent Putman]

On 6/5/12 3:48 PM, Chad La Joie wrote:
>
> Yeah, I checked the spec, integer is in fact defined to be the infinite
> set of all whole numbers.  Which leads to the idiosyncratic definition
> of 'long' being a smaller set of numbers (-9223372036854775808 -
> 9223372036854775807).

I had forgotten that there is (confusingly) both xsd:integer and
xsd:int. The former is infinite, the latter is the equivalent of a
32-bit int from the common languages.  So at least in that sense
xsd:long and xsd:int are what one would expect relative to one another. 
xsd:integer is the odd one.


> In OSv3, I recommend we move to longs.  It's not "infinite", and it'll
> probably bite us at some point, but things like BigInteger just seem
> like a bad idea for this.

That was my initial gut feeling as well. However I did some checking,
and for example both JAXB and XmlBeans do by default use BigInteger for
representing xsd:integer  (and int/Integer or long/Long for xsd:int and
xsd:long respectively).  So there is at least some precedent for it.

I was kinda hoping that there might be some library with an arbitrary
precision integer rep in it that we could use, but haven't found one
yet.  I think we should at least think on it some more.  It's
intellectually somewhat hard to defend using Long since it's not any
more correct than Integer.