Function: completion--try-word-completion

completion--try-word-completion is a byte-compiled function defined in minibuffer.el.gz.

Signature

(completion--try-word-completion STRING TABLE PREDICATE POINT MD)

Source Code

;; Defined in /usr/src/emacs/lisp/minibuffer.el.gz
(defun completion--try-word-completion (string table predicate point md)
  (let ((comp (completion-try-completion string table predicate point md)))
    (if (not (consp comp))
        comp

      ;; If completion finds next char not unique,
      ;; consider adding a space or a hyphen.
      (when (= (length string) (length (car comp)))
        ;; Mark the added char with the `completion-word' property, so it
        ;; can be handled specially by completion styles such as
        ;; partial-completion.
        ;; We used to remove `partial-completion' from completion-styles
        ;; instead, but it was too blunt, leading to situations where SPC
        ;; was the only insertable char at point but minibuffer-complete-word
        ;; refused inserting it.
        (let ((exts (mapcar (lambda (str) (propertize str 'completion-try-word t))
                            '(" " "-")))
              (before (substring string 0 point))
              (after (substring string point))
	      tem)
          ;; If both " " and "-" lead to completions, prefer " " so SPC behaves
          ;; a bit more like a self-inserting key (bug#17375).
	  (while (and exts (not (consp tem)))
            (setq tem (completion-try-completion
		       (concat before (pop exts) after)
		       table predicate (1+ point) md)))
	  (if (consp tem) (setq comp tem))))

      ;; Completing a single word is actually more difficult than completing
      ;; as much as possible, because we first have to find the "current
      ;; position" in `completion' in order to find the end of the word
      ;; we're completing.  Normally, `string' is a prefix of `completion',
      ;; which makes it trivial to find the position, but with fancier
      ;; completion (plus env-var expansion, ...) `completion' might not
      ;; look anything like `string' at all.
      (let* ((comppoint (cdr comp))
	     (completion (car comp))
	     (before (substring string 0 point))
	     (combined (concat before "\n" completion)))
        ;; Find in completion the longest text that was right before point.
        (when (string-match "\\(.+\\)\n.*?\\1" combined)
          (let* ((prefix (match-string 1 before))
                 ;; We used non-greedy match to make `rem' as long as possible.
                 (rem (substring combined (match-end 0)))
                 ;; Find in the remainder of completion the longest text
                 ;; that was right after point.
                 (after (substring string point))
                 (suffix (if (string-match "\\`\\(.+\\).*\n.*\\1"
                                           (concat after "\n" rem))
                             (match-string 1 after))))
            ;; The general idea is to try and guess what text was inserted
            ;; at point by the completion.  Problem is: if we guess wrong,
            ;; we may end up treating as "added by completion" text that was
            ;; actually painfully typed by the user.  So if we then cut
            ;; after the first word, we may throw away things the
            ;; user wrote.  So let's try to be as conservative as possible:
            ;; only cut after the first word, if we're reasonably sure that
            ;; our guess is correct.
            ;; Note: a quick survey on emacs-devel seemed to indicate that
            ;; nobody actually cares about the "word-at-a-time" feature of
            ;; minibuffer-complete-word, whose real raison-d'être is that it
            ;; tries to add "-" or " ".  One more reason to only cut after
            ;; the first word, if we're really sure we're right.
            (when (and (or suffix (zerop (length after)))
                       (string-match (concat
                                      ;; Make submatch 1 as small as possible
                                      ;; to reduce the risk of cutting
                                      ;; valuable text.
                                      ".*" (regexp-quote prefix) "\\(.*?\\)"
                                      (if suffix (regexp-quote suffix) "\\'"))
                                     completion)
                       ;; The new point in `completion' should also be just
                       ;; before the suffix, otherwise something more complex
                       ;; is going on, and we're not sure where we are.
                       (eq (match-end 1) comppoint)
                       ;; (match-beginning 1)..comppoint is now the stretch
                       ;; of text in `completion' that was completed at point.
		       (string-match "\\W" completion (match-beginning 1))
		       ;; Is there really something to cut?
		       (> comppoint (match-end 0)))
              ;; Cut after the first word.
              (let ((cutpos (match-end 0)))
                (setq completion (concat (substring completion 0 cutpos)
                                         (substring completion comppoint)))
                (setq comppoint cutpos)))))

	(cons completion comppoint)))))