(encoded-kbd-self-insert-utf-8): Catch and recover from case when the bytes

we thought we were reading turn out to be something else entirely, such as
latin-1 chars from quail.  See bug#396.

2 files changed, 26 insertions, 4 deletions

diff --git a/lisp/ChangeLog b/lisp/ChangeLog
index fd6246c5916..42b7c94c6aa 100644
--- a/lisp/ChangeLog
+++ b/lisp/ChangeLog
@@ -1,3 +1,10 @@
+2008-06-15  Stefan Monnier  <monnier@iro.umontreal.ca>
+
+        * international/encoded-kb.el (encoded-kbd-self-insert-utf-8):
+        Catch and recover from case when the bytes we thought we were reading
+        turn out to be something else entirely, such as latin-1 chars from
+        quail.  See bug#396.
+
 2008-06-15  Dan Nicolaescu  <dann@ics.uci.edu>
 
         * vc.el (vc-deduce-fileset): Check if the buffer has a file.
diff --git a/lisp/international/encoded-kb.el b/lisp/international/encoded-kb.el
index ec887659e9c..4659b499112 100644
--- a/lisp/international/encoded-kb.el
+++ b/lisp/international/encoded-kb.el
@@ -219,8 +219,9 @@ The following key sequence may cause multilingual text insertion."
 
 (defun encoded-kbd-self-insert-utf-8 (arg)
   (interactive "p")
-  (let ((char (encoded-kbd-last-key))
-        len)
+  (let* ((lead (encoded-kbd-last-key))
+         (char lead)
+         len event)
     (cond ((< char #xE0)
            (setq len 1 char (logand char #x1F)))
           ((< char #xF0)
@@ -230,8 +231,22 @@ The following key sequence may cause multilingual text insertion."
           (t
            (setq len 4 char 0)))
     (while (> len 0)
-      (setq char (logior (lsh char 6) (logand (read-char-exclusive) #x3F))
-            len (1- len)))
+      (setq event (read-char-exclusive))
+      (if (and (>= event #x80) (< event #xc0))
+          ;; Valid utf-8 sequence.
+          (setq char (logior (lsh char 6) (- event #x80))
+                len (1- len))
+        ;; Invalid utf-8 sequence.  Might be because Quail got involved
+        ;; in-between and the bytes we thought we were reading were actually
+        ;; latin-1 chars.  Let's presume that `event' is the second "byte",
+        ;; i.e. there weren't any "apprently correct" between `lead' and
+        ;; `event': it's easy to recover in this case, and the more general
+        ;; case seems pretty unlikely.
+        ;; FIXME: We should really do encoded-kbd decoding before processing
+        ;; input-methods.
+        (push event unread-command-events)
+        (setq char lead)
+        (setq len 0)))
     (vector char)))
 
 (defun encoded-kbd-setup-keymap (keymap coding)