bug#17239: 24.3.50; competion error (cl-assertion-failed (string-prefix-p uprefix ustring))

[Stefan Monnier]

> C-x C-f
> / C-b TAB ;; hit tab between the last two slashes.
> => error.

> Debugger entered--Lisp error: (cl-assertion-failed (string-prefix-p uprefix 
> ustring))

Thanks.  I installed the patch below which fixes the most glaring
problem (the assertion failure).  The resulting behavior is not great,
but it's not really clear what we should do anyway.


        Stefan


=== modified file 'lisp/ChangeLog'
--- lisp/ChangeLog      2014-05-03 08:47:50 +0000
+++ lisp/ChangeLog      2014-05-05 01:44:49 +0000
@@ -1,3 +1,8 @@
+2014-05-05  Stefan Monnier  <monnier@iro.umontreal.ca>
+
+       * minibuffer.el (completion-table-with-quoting) <completion--unquote>:
+       Make sure the new point we return is within the new string (bug#17239).
+
 2014-05-03  Eli Zaretskii  <eliz@gnu.org>
 
        * mail/rmailsum.el (rmail-new-summary-1): Fix a typo in a comment.

=== modified file 'lisp/minibuffer.el'
--- lisp/minibuffer.el  2014-05-01 23:55:25 +0000
+++ lisp/minibuffer.el  2014-05-05 01:24:34 +0000
@@ -519,11 +519,35 @@
         completions))
 
      ((eq action 'completion--unquote)
-      (let ((ustring (funcall unquote string))
-            (uprefix (funcall unquote (substring string 0 pred))))
-        ;; We presume (more or less) that `concat' and `unquote' commute.
-        (cl-assert (string-prefix-p uprefix ustring))
-        (list ustring table (length uprefix)
+      ;; PRED is really a POINT in STRING.
+      ;; We should return a new set (STRING TABLE POINT REQUOTE)
+      ;; where STRING is a new (unquoted) STRING to match against the new TABLE
+      ;; using a new POINT inside it, and REQUOTE is a requoting function which
+      ;; should reverse the unquoting, (i.e. it receives the completion result
+      ;; of using the new TABLE and should turn it into the corresponding
+      ;; quoted result).
+      (let* ((qpos pred)
+            (ustring (funcall unquote string))
+            (uprefix (funcall unquote (substring string 0 qpos)))
+            ;; FIXME: we really should pass `qpos' to `unuote' and have that
+            ;; function give us the corresponding `uqpos'.  But for now we
+            ;; presume (more or less) that `concat' and `unquote' commute.
+            (uqpos (if (string-prefix-p uprefix ustring)
+                       ;; Yay!!  They do seem to commute!
+                       (length uprefix)
+                     ;; They don't commute this time!  :-(
+                     ;; Maybe qpos is in some text that disappears in the
+                     ;; ustring (bug#17239).  Let's try a second chance guess.
+                     (let ((usuffix (funcall unquote (substring string qpos))))
+                       (if (string-suffix-p usuffix ustring)
+                           ;; Yay!!  They still "commute" in a sense!
+                           (- (length ustring) (length usuffix))
+                         ;; Still no luck!  Let's just choose *some* position
+                         ;; within ustring.
+                         (/ (+ (min (length uprefix) (length ustring))
+                               (max (- (length ustring) (length usuffix)) 0))
+                            2))))))
+        (list ustring table uqpos
               (lambda (unquoted-result op)
                 (pcase op
                   (1 ;;try
@@ -853,6 +877,7 @@
              (setq string (pop new))
              (setq table (pop new))
              (setq point (pop new))
+            (cl-assert (<= point (length string)))
              (pop new))))
         (result
          (completion--some (lambda (style)