string-lessp uses string_char() to access individual charactrs, even under Mule where that operation is O(n). This makes string comparison O(n^2), potentially very slow for large strings. This is not noticable with short strings, but is catastrophic with longer ones (and I can only guess what it does to sorting). To test the slowness of comparing long strings, try this (under Mule): (setq s1 (make-string 10000 ?a)) <C-x C-e> (setq s2 (make-string 10000 ?a)) <C-x C-e> (string-lessp s1 s2) With this patch the comparison becomes instantaneous: 1999-06-04 Hrvoje Niksic <hniksic(a)srce.hr&gt; * fns.c (Fstring_lessp): Remove O(n^2) under Mule.