[antlr-interest] [C target] [3.1.1] Deriving imaginary nodes from real tokens

[Gavin Lambert]

At 01:52 13/12/2008, Sven Van Echelpoel wrote:
 >Ah, I wondered why there were two forms of the template. Can you 

 >really pass constructor arguments to nodes created from token
 >references?  Why would you want to use the info of another token 

 >(I can buy the text), as in the following 2 argument call:
 >
 >-> $b[ $c, "c" ]

I think you've misinterpreted what I was saying.  I'm fairly sure 
that what you've posted there is *not* legal (it doesn't make 
sense to pass two tokens into one construction expression).

As far as I know, the following are the only valid token 
construction expressions (I could be wrong about some of the 
specifics, though):

   -> FOO
Copy existing token FOO, or create imaginary token FOO with no 
context.

   -> FOO["bar"]
Copy existing token FOO but give it "bar" as text, or create 
imaginary token FOO with "bar" as text and no context.

   -> FOO[$b]
Copy existing token in $b but change the type to FOO.

   -> FOO[$b, "bar"]
Copy existing token in $b but change the type to FOO and the text 
to "bar".

   -> $b
Copy existing token in $b.

   -> $b["bar"]
Copy existing token in $b, but change the text to "bar".

(And unfortunately, the C target can't distinguish between #2 and 
#3 because the expression can be such that it can't statically 
tell whether it returns a string or a token.)