Re: string-match doesn't honor anchor when START is nonzero

[Jim Meyering <meyering@ascend.com>]

Richard Stallman <rms@gnu.org> writes:

|     Is it expected that this would evaluate to nil?  (it does)
|
|       (string-match "^j" " j" 1)
|
| That is correct.  The argument 1 means to look for a match starting no
| earlier than position 1, but the criterion for a match does not treat
| position 1 as the start of a line if it is not one.
|
|     If returning `nil' is the intended behavior, then there's a bug in
|     mail-utils.el (rmail-dont-reply-to) because it seems to expect `1' and
|     fails when it gets the `nil'.
|
| Could you show which call to string-match is seems to have the problem?
| It seems you have tracked this down already; I did a quick scan and
| I don't see a bug of this kind.  We could debug it, but if you already
| have done so, why not tell us what you know?

  This illustrates the problem:

    (setq rmail-dont-reply-to-names "f.")
    (load "mail-utils")
    (rmail-dont-reply-to "b, f1, f2, f3, f4, f5, f6")

    Return value: "b, f2, f4, f6"

  It should be

    Return value: "b"

Here's part of mail-utils.el:
========================================
(defun rmail-dont-reply-to (userids)
  "Returns string of mail addresses USERIDS sans any recipients
that start with matches for `rmail-dont-reply-to-names'.
Usenet paths ending in an element that matches are removed also."
  (if (null rmail-dont-reply-to-names)
      (setq rmail-dont-reply-to-names
	    (concat (if rmail-default-dont-reply-to-names
			(concat rmail-default-dont-reply-to-names "\\|")
		        "")
		    (concat (regexp-quote (user-login-name))
			    "\\>"))))
  (let ((match (concat "\\(^\\|,\\)[ \t\n]*"
		       ;; Can anyone figure out what this is for?
		       ;; Is it an obsolete remnant of another way of
		       ;; handling Foo Bar <foo@machine>?
		       "\\([^,\n]*[!<]\\|\\)"
		       "\\("
			     rmail-dont-reply-to-names
		       "\\|"
		             ;; Include the human name that precedes <foo@bar>.
			     "\\([^\,.<\"]\\|\"[^\"]*\"\\)*"
			     "<\\(" rmail-dont-reply-to-names "\\)"
		       "\\)"))
	(case-fold-search t)
	pos epos)
    (while (setq pos (string-match match userids pos))
      (if (> pos 0) (setq pos (match-beginning 2)))
      (setq epos
	    ;; Delete thru the next comma, plus whitespace after.
	    (if (string-match ",[ \t\n]*" userids (match-end 0))
		(match-end 0)
	      (length userids)))
========================================

This part of the regex

  "\\(^\\|,\\)[ \t\n]*"

is wrong the second time around because we've already advanced past the
comma (2nd string-match above).  That means we end up advancing past a
potentially matching address to the next comma.  Hence, in the example
above, only every other matching address is removed.