Parse in boundaries

[volrath]

I'm currently in the need of parsing just a fraction of a buffer, so I thought on changing the signature of parseclj-parse-clojure to something like:

(defun parseclj-parse-clojure (&optional bounds-or-string options)
  ...)

bounds-or-string can be either a list (or cons) with positions/markers, a string that will be put into a temp buffer and scanned from (point-min) to (point-max), or nil, in which case just scans from (point) til the end (current default behavior)

I wanted to hear your thoughts on it before implementing.

My temporary solution is to use narrowing:

(save-excursion
  (goto-char (car bounds))
  (apply #'narrow-to-region bounds)
  (prog1 (parseclj-parse-clojure)
    (widen)))

But I think it's not ideal.

I was going to suggest narrowing but I see you already thought of that. That would be my recommended solution. What do you dislike about it? 

[volrath]

Basically the save-excursion. I've read it's costly in terms of performance, and since we're going for high performance here...

On the other hand I also get that we'd need to also change parseclj-parser signature to get the boundaries so that the shift-reduce process would stop at either :eof or (cdr bounds), which also mean tweak with parseedn...

[plexus]

What do you think of doing some benchmarks on this? I can see the need for this behavior, but it'll add a good bit of complexity, so it would be good to know it'll be worth it...

Things we could compare to get a base line:

• current behavior of just scanning a buffer until the end
• use save-excursion and narrow-to-region
• use buffer-substring (or buffer-substring-no-properties, I'm not sure which one is actually faster), and pass that to the parser (which then creates a temp buffer).

If the second two are really a lot slower than the first one, then it's worth it to change the API. In that case I would prefer making it two options, :start which defaults to (point), and :end which defaults to (point-max). I like that better than having to pass in an extra nil just to get the base behaviour.

[volrath]

Both things (benchmarking and using :start/:end options) sound good to me :)

[MalloZup]

autogenerated with https://github.com/MalloZup/doghub: issue inactive since 450 days. Please update the issue or close it