Handling "undeclared" references

[Frans Englich]

Hi everyone,

I have a parsing problem which I don't know how to solve the best way. In 
short, it is that I need to parse XML.[1] The problem's essence is shown in 
this example:

<prefix:name attr1="content" prefix:attr2="content2" 
xmlns:prefix="example.org/HereIsPrefixDeclared"/>

The problem is that "xmlns:prefix" declares the prefix 'prefix', but it is 
used before it is declared. It's used in "prefix:name" and "prefix:attr2". In 
other words, this is a problem in the same direction of the example in 
"Actions in Mid-Rule"(bison manual), but a bit more tricky.

In my case, I find the attributes simple to handle. I retrieve all the 
attributes and they don't need any namespace declarations because I deal with 
it manually in the action.

However, the start tag, "prefix:name", is a bit more problematic. I could of 
course just deal with it manually, but I would really want to push it down 
into an action, and have it dealt with there. Here's how I would like to do: 
to let the parser tokenize and see what rules to use, but to wait with 
invoking the actual action for the rule handling "<prefix:name", until all 
the attributes are parsed. Since then the namespace declarations(attributes) 
have populated my parse "context" with declarations. Is anything in that 
direction possible?

Another solution is to add a layer of indirection. Instead of letting the 
parser resolve namespace declarations, I use "place holders" in my AST, and 
let an AST-pass do a "fixup". But that surely is heavy and not very elegant, 
if you ask me.

Comments welcome. I would gladly hear about how people solve cases like this.


Thanks in advance,

                Frans

1.
Well, not XML 1.0 to be exact, but XML literals occuring in the XQuery 
language. However, the problem is identical to as if constructing a parser 
for XML.