Many thanks for the advice ! I've start forging ahead with the syntax-property approach since that seems the most robust and provides the least coupling between the stages.
On Thu, Apr 4, 2013 at 2:39 AM, Sam Tobin-Hochstadt <sa...@ccs.neu.edu>wrote: > On Wed, Apr 3, 2013 at 10:31 PM, Robby Findler > <ro...@eecs.northwestern.edu> wrote: > > > > You can put syntax properties on the result of your expander for the > > state-machine macro and then search for those properties in the expanded > > text. Maybe that will help? > > There's information about how Typed Racket does this (using roughly > the technique Robby outlined, plus others) in our papers about the > implementation, from PLDI 2011 and the Scheme Workshop 2007 [1,2]. > > The other significant technique is basically the following: > > Expand a form like: > > (my-special-dsl some arguments here) > > to > > (begin (define-values () (begin (quote-syntax (my-special-dsl some > arguments here)) (values)) ;; just the original input syntax > the-actual-expansion) > > Then your tool can look at the results of expansion (via the hook > Robby describes) and find occurrences of this pattern, and discover > your DSL. Note that this only really works if either (a) your DSL > doesn't have any general expansion positions as subforms or (b) you > don't need to analyze those subforms. Otherwise syntax properties are > the best route. > > Sam > > [1] http://www.ccs.neu.edu/racket/pubs/scheme2007-ctf.pdf > [2] http://www.ccs.neu.edu/racket/pubs/pldi11-thacff.pdf >
____________________ Racket Users list: http://lists.racket-lang.org/users
