On Sun, Mar 08, 2020 at 08:10:50AM -0700, Matt Wette wrote:
On 3/8/20 3:14 AM, address@hidden wrote:
Hi,
[...]
My question is: where is the stuff resolved which is mentioned
in grammar actions?
Hah. Managed to answer my own question by reading the source.
For the benefits of others who might have a similar question
(without knowing how to articulate it, like it happened to
me), here's the answer:
Yes, the actions are "resolved" wrt the (calling) module's
top level environment. The magic happens here (that's wrt the
all-fresh V1.01.2), around lines 47 ff, in module/nyacc/parse.scm:
(define (make-xct av)
(if (procedure? (vector-ref av 0))
av
(vector-map (lambda (ix f) (eval f (current-module)))
(vector-map (lambda (ix actn) (wrap-action actn)) av))))
It's the (eval f (current-module)) which does the trick. The
trick happens in make-lalr-parser.
Background: I envisioned something like
(let ( ...some environment for the actions...)
...make-lalr-parser...)
and have make-lalr-parser pick up the bindings in the lexical
environment.
but had to realize that make-lalr-parser ignores the lexical
environment. That's now clear to me, because (eval ... (current-module))
looks at the caller's module's top-level bindings.
One would have to call local-eval (from (ice-9 local-eval)) and
explicitly pass (the-environment) (from the same module) to
achieve what I had in mind.
Not that I'm proposing that, mind you. I still barely know what
I'm doing at this point. Probably there are very good reasons
to resort to the (top-level) module bindings.
I just wanted to understand, and I think I do now :-)
As to your other proposal (writing out the pre-compiled parser
with write-lalr-*), I think it's orthogonal to my issue. This
might come in handy when the parser is heavy enough that it
takes a significant time constructing it (which it is not in my
little use case for now)
Thanks & cheers
-- tomás