bug#35708: [27.0.50]: thingatpt.el, thing-at-point-looking-at redundant

[npostavs]

Andreas Röhler <andreas.roehler@easy-emacs.de> writes:

>> Hmm, current thing-at-point-looking-at might be slow with large
>> buffers. The slightly modified test should reveal it:
>>
>> (ert-deftest thing-at-point-looking-at-2 ()
>>   (with-temp-buffer
>>     (insert "1abcd 222abcd")
>>     (dotimes (_ 99999) (insert " asdf "))
>>     (goto-char (point-min))

>>       (search-forward "2ab")
>>       (should (thing-at-point-looking-at "2abcd"))

Yes, in this case, since the loop over looking-at only needs to iterate
twice, so it will be faster.  But what about when there is no match?
E.g.,

(with-temp-buffer
  (dotimes (_ 99999) (insert " asdf "))
  (goto-char (point-max))
  (list :ar-regexp-atpt (benchmark-run (ar-regexp-atpt "foo"))
        :thing-at-point-looking-at (benchmark-run (thing-at-point-looking-at 
"foo"))))

> Another fix, as a bug showed up when testing (ar-regexp-atpt "[a-z]+"):
>
> (defun ar-regexp-atpt (regexp)
>   "Return t if REGEXP matches at or before point, nil otherwise.
>
> Changes match-data"
>   (save-excursion
>     (if (looking-at regexp)
>     (while
>         (and (not (bobp))
>          (or (progn (backward-char) (looking-at regexp))
>              (forward-char 1))))
>       (while (not (or (bobp) (backward-char) (looking-at regexp))))
>       (ar-regexp-atpt regexp))

What's this recursive call for?  It triggers (error "Lisp nesting
exceeds ‘max-lisp-eval-depth’") in the benchmark above.

>     (looking-at regexp)))