(<?xml version="1.0"
encoding="UTF-8"?>,"foo",5^positiveInteger,<bar><bat/></bar>,<!--
foo -->)
I'm not at all opposed to multiple new serialization formats, although
I'm inclined to think getting *one* more with any decent adoption is a
ambitious goal, let alone 2.
I figure in for a penny, in for a pound. Sequence serialization, with or without XDM, is a problem that will only become more prevalent as we start to use sequences of documents more routinely. I'm thinking especially here of the XQuery and XProc use-cases, both of which are fully capable of generating sequences. Changing serialization formats at this stage is a lot easier because of relatively early adoption of these technologies, but that will change as both become more commonplace.
Your example with RNG is interesting, but I don't think its quite a
parallel. With RNG the non-xml form is intended to be authored by
humans, with a design goal of simple human editable representations.
In this case, so far none of the design goals (or use cases) I've come
up with yet involve humans authoring the data.
In your example, what is the design intention for a non-xml format ?
I'm not necessarily saying that there is one, only that if you introduce a formal xdm: or xml: notation for handling serialization/deserialization, this will then need to be processed in some manner in order to create the internal sequence representation, which implies a pre-process step of some sort, whether xml or not. The non-xml format has the advantage of compactness - for some people this is a consideration (for me it probably wouldn't be, but there are people for whom this is a big factor).
In my mind, there is one example where non-xml format for sequences
would be very useful but I'm not satisfied with how it would actually
work in practice.
that is, I believe the most common actual production of XDM data
happens to be either plain text, or a single XML item (element,
document).
In both of those cases it would be really nice if the serialization
happened to be the 'standard' serialization for those without any kind
of wrapping at all,
(no ( ) or no <xdm:wrapper> .. etc)
That way if you just happened to produce a single XDM Item of type
element or text there'd be no extra baggage.
I think that would be really cool. But the only way I've thought of
to achieve that is to use a sequence delimited format with no start and
end markers.
I don't really think that singleton content is that much of an issue. You can serialize most singletons now without needing any additional content, but it's worth noting here that such content is, by definition, scalar and dimensionless. If you had an xdm serialization, it actually might make sense to have an <xdm:wrapper> around such singletons, if only because this could be used to provide type information. There is also a distinct difference between a naked singleton and singleton entry of a sequence - the former would just be the xml representation (with or without the encoding header), the latter would be an xdm:wrapper (or xdml:sequence) element surrounding the sequence itself.
My opinion is that if I'm going to have to parse "(" and "," I'd rather
be parsing "<wrapper> ... </wrapper>" at least I wouldn't
have to write a new (if even simple) parser and can simply read it as
XML. For example I would like to provide a 'sample implementation' of
the serialize and parser written in pure XQuery as an additional way of
describing the format besides prose.
But perhaps your thinking of a use case or design goal I have neglected.
Not really. Either way, you'd have to define an EXPath (or fill in the blank standard) set of xdm:serialize() and xdm:parse() in order to track into internal XDM representations. The XML representations are complicated only by the fact that there is no consistent serialization or parse mechanisms in the fn: namepace (eXist's declare-option function would be the closest (and I think there's something analogous in Mark Logic)) but otherwise you'd have to manually walk the tree for each serialization.
Kurt
