bug#54733: Match again in perform-replace

[Juri Linkov <juri@jurta.org>]

>>> +				(replace-search search-string limit
>>> +						regexp-flag delimited-flag
>>> +						case-fold-search backward))
>>
>> I don't know this code too well, but perhaps SEARCH_STRING here should be
>> anchored with something like "\\=" at the beginning?
>>
>> Otherwise the search can succeed here even if the next match is not
>> here. Not sure how important that is, though.
>
> I'm afraid that prepending "\\=" might break a complex regexp somehow.

Another variant without modifying the original regexp is still to call
replace-search, but afterwards check if it stayed at the old position
with something like

  (let ((old-pos (point)))
    (and (replace-search ...)
	 (eq (match-beginning 0) old-pos)))

Less efficient, but looking-at is a real problem since it's incompatible
with search functions.  This is one of the problems faced in bug#14013
where isearch matches `C-M-r ^' outside of positions handled by the
search function because `isearch-search-and-update' uses a hack with
looking-at.  This is a long-standing flaw in isearch that needs to be fixed.
Any uses of looking-at in search/replace should be substituted
with an equivalent code that relies on the search function.