geoff wrote: [re LoopDouble accuracy]
I think that this issue is another implementation concern for which we should only provide recommendations in the specification, and agree that this should not affect the schema (nor lead us to consider mandating a particular language).
Non-normative recommendations are a good plan. [re LoopDouble formatting]
For this issue I'm again (perhaps unsurprisingly! :) ) convinced that it is an implementation concern.
An XML parser/validator will naturally be expected to validate that a JSDL XML document conforms to the defined schema, e.g. ensure specified values are of type xsd:decimal, xsd:double, etc.. Beyond that though, in terms of subsequent processing, I very much think that we should only provide recommendations where appropriate. Even if we specified formats we cannot mandate underlying implementations to use them. Using the example above I personally would only anticipate recommending format consistency when specifying looping.
I suspect this is something that we may have to revisit in the future, though it is only a real problem when you're substituting the numbers into a context like a fragment of a filename, where you're more likely to have a "number-like thing" than a real number. When it is going into a control file, it probably isn't a problem in practice. Could be that my feel for this issue isn't perfect. It's probably a good thing to recommend that implementors keep the format consistent. Not good to flip back and forth between different notation types. (Yes, non-normative for that will do.) [re equality tests for doubles in Exception elements]
Once more I feel we should only recommend to implementors that they may wish to make such calculations.
That'd be fine as a non-normative recommendation. Was just suggesting things that might make it easier to write. :-) Donal.