Support imenu-like feature about emacs HOT 7 OPEN

I duplicated the buffer content a bunch of times and got a 1k line buffer, then tried the code again. It is still pretty fast, took about 0.1 sec. I think we can probably get away with this approach :-) Add an idle-timer, while-no-input and double-buffering, and it should be smooth.

from emacs.

I will investigate if your code is faster than the dynamic module option. I mean, it almost certainly is. However, the example I gave is of course just a simplified version of the actual tree generator which is primarily -- but not only -- designed for echo area navigational aid as you traverse your code. See:

Now this is the hacky version; the original rendered a proper subset of the sparse tree but that of course (with ETS) turned out to have some performance bottlenecks. This simplified version is probably 'good enough' though.

from emacs.

Ah, I see, very interesting feature indeed. I'll wait for you result. If it is still not fast enough, you can describe what you do in more detail, and hopefully we can come up with something in C that can boost the performance for this type use use case.

from emacs.

while I'm excited to see tree-sitter getting integrated into the Emacs core, I'd very much like the default imenu integration to be synchronous (but fast) and not do anything in the background that might lead to ~100ms latency spikes.
I'm currently using a primitive custom imenu definition that uses the tree-sitter dynamic module to gather a simple list of method names; it's not as pretty as the hierarchical structure shown here, so having both to choose from would be the ideal solution IMO, but it's quick (~3 to 5 ms on files in the 1K line ballpark).

Note that I'm not currently trying to extract scopes, and I'm not sure if that can be done within tree-sitter's query mechanism or if one would have to manually traverse the AST starting from matched nodes (assuming that is supported by the API). Would that be a major performance bottleneck?

For completeness, this is the primitive implementation I'm using (hence the sebastian/ prefix, it's part of my private config):

(defmacro sebastian/define-tsc-index-function (fn-name query)
  `(defun ,fn-name ()
     (unless sebastian/tsc--imenu-query
       (setq sebastian/tsc--imenu-query (tsc-make-query tree-sitter-language ,query)))
     (tsc--without-restriction
      (let* ((matches
              (tsc-query-matches
               sebastian/tsc--imenu-query
               (tsc-root-node tree-sitter-tree) #'tsc--buffer-substring-no-properties)))
        (mapcar (lambda (match)
                  (pcase (cdr match)
                    (`[(name . ,name) (pars . ,pars)]
                     (cons
                      (string-replace
                       "( " "("
                       (s-collapse-whitespace
                        (string-join (list (tsc-node-text name) (tsc-node-text pars)))))
                      (tsc-node-start-position name)))))
                matches)))))

;; for C++ buffers:
(sebastian/define-tsc-index-function
 nq-cc-index-function
 [(function_declarator
   [(operator_name) (field_identifier) (scoped_identifier) (identifier)] @name
   (parameter_list) @pars)])

  (setq imenu-create-index-function #'nq-cc-index-function)

from emacs.

Yes, if we do provide a default tree-sitter imenu integration, it would be like what you showed here. I don't think imenu support scope anyway.

from emacs.

I mean, imenu was just an example, of course. It should be blazing fast; indeed, it all should, within the limits of what we can do with Emacs and so forth.

from emacs.

glad to hear that, no reason for me to worry then :)

from emacs.