bug#37321: 27.0.50; Excessive gc in a use case (el-search)

[Michael Heerdegen <michael_heerdegen@web.de>]

Michael Heerdegen <michael_heerdegen@web.de> writes:

> M-x el-search-emacs-elisp-sources 'pcase M-RET

I found the culprit.  It's this simple function:

#+begin_src emacs-lisp
(defun el-search--flatten-tree (tree)
  "Return a list of `el-search--atomic-p' objects in TREE."
  (let ((elements ())
        (walked-objects ;to avoid infinite recursion for circular TREEs
         (make-hash-table :test #'eq))
         (gc-cons-percentage 0.8)
        ) ;Why is binding it here more effective than binding it more top-level?
    (cl-labels ((walker (object)
                        (if (el-search--atomic-p object)
                            (push object elements)
                          (unless (gethash object walked-objects)
                            (puthash object t walked-objects)
                            (if (consp object)
                                (progn
                                  (while (consp object)
                                    (walker (car object))
                                    (setq object (cdr object))
                                    (if (gethash object walked-objects)
                                        (setq object nil)
                                      (puthash object t walked-objects)))
                                  (when object ;dotted list
                                    (walker object)))
                              (cl-loop for elt being the elements of object do (walker elt)))))))
      (walker tree)
      elements)))
#+end_src

I've put a lot of work in optimizing el-searching speed.  It now mainly
depends of effectiveness of gc, mostly in the above function.  When I
wrote it I noticed that it worked best when binding gc-cons-percentage
to a value of 0.8 (inside the function, see the code).  But now this
binding slows the thing down as hell.  When I remove the binding,
searching is fast again, but it still takes twice as long as in the past
when 0.8 worked.

I've no clue about how gc works and how the above function can be
optimized, and if gc is expected to be that slow now with the
gc-cons-percentage binding in effect.  Advice appreciated.

TIA,

Michael.