[XML-SIG] RAX

[Ken MacLeod]

Lars Marius Garshol <larsga@garshol.priv.no> writes:

> * Ken MacLeod
> | 
> | Since that's good for a reason, it may also be good if there were a
> | version of SAX that _was_ pull-style so that it could be used in
> | applications like RAX.  In this way, one could stack SAX modules and
> | filters in a pull-style chain as an alternative to a push-style chain.
> 
> I agree with this and have been thinking about this for a while, but
> I'm not sure how we would actually implement this. The only XML parser
> we have that supports a pull-style interface is RXP, and I'm not sure
> if we can convert the other interfaces to pull-style interfaces in a
> sensible way (at least not on a level as low as SAX) without storing
> the entire sequence of events.
> 
> Good ideas are welcome...

I don't think existing push-style parsers need to be converted, or
implied that they could be used in a pull-style chain.  I was thinking
more of creating the interface definition of a pull-style SAX parser
and allowing for new parsers to be developed rather than a wholesale
conversion of push-style parsers.

RXP and PYX are both good candidates for pull-style parsing.

I think an EasySAX-like approach would work best, where next_event()
returns a DOM/mini-DOM node:

    node, is_end = pull_parser.next()
    while node != None:
        if node.nodeType == ELEMENT:
            if is_end:
                """ do end element processing """
            else:
                """ do start element processing """
        elif node.nodeType == TEXT:
            """ do text processing """

        node, is_end = pull_parser.next()

Most of the rest of the SAX interface (sources, creating parsers,
exceptions, locators) could probably be used without change.

If two threads are used, any push-style parser can be used to queue
events to be read by a pull-style adapter in the other thread.

  -- Ken