bug#49534: 26.3; Isearch should support using filter predicates with empty search hits

[Juri Linkov <juri@linkov.net>]

>>>  (= (match-beginning 0) (match-end 0))
>>
>> And, indeed, the default predicate doesn't match on invisible text...
>> but I'm not sure why it's also testing the length of the match.  (The
>> default predicate also checks this, so removing the test seems to
>> produce identical results by default.)
>
> I've tried to remove (= (match-beginning 0) (match-end 0))
> and then tried the test case provided by Drew,
> but it goes into an infinite loop.
>
> So it requires advancing by 1 char - the same trick as it's used
> in query-replace, etc.

Oh, I realized that

  (= (match-beginning 0) (match-end 0))

was moved below after

  (funcall isearch-filter-predicate
           (match-beginning 0) (match-end 0))

This means that if isearch-filter-predicate does own matching,
it will break later (= (match-beginning 0) (match-end 0)).

What would be better: to remember its result in a let-bound variable,
or to use save-match-data?  Probably, save-match-data:

diff --git a/lisp/isearch.el b/lisp/isearch.el
index 3337d9be68..9113e94c3b 100644
--- a/lisp/isearch.el
+++ b/lisp/isearch.el
@@ -3591,8 +3591,9 @@ isearch-search
 	  ;; Clear RETRY unless the search predicate says
 	  ;; to skip this search hit.
 	  (if (or (not isearch-success)
-		  (funcall isearch-filter-predicate
-			   (match-beginning 0) (match-end 0)))
+		  (save-match-data
+                    (funcall isearch-filter-predicate
+			     (match-beginning 0) (match-end 0))))
 	      (setq retry nil)
 	    ;; Advance point on empty matches before retrying
 	    (when (= (match-beginning 0) (match-end 0))