This patch to elogapply() silences "elogapply: can't happen ..." messages
that occur under the following somewhat rare circumstance:
  1. a text selection near the end of file,
  2. a command is executed that replaces the selected text, and
  3. the length of the replacement text is sufficiently smaller than the selection.
This can result in the saved 'q0'/'q1' value exceeding the new buffer cache length,
even though the textdelete call did the right thing. 

The proposed change updates 'q0' and 'q1' after the call to textdelete under "case
 Delete".  It doesn't appear to break anything, but I am by no means a 'power'
acme user, so further investigation would be prudent. "Case Replace" may also
require some fiddling but I couldn't figure out how to exercise that case.

Notes:
Wed Mar 23 14:41:57 EST 2005 rsc
    Thank you for sending instructions to reproduce this problem!
    I've never been able to do that.
    
    Your patch treated the symptom instead of the actual disease.
    I finally understand the disease and I think I've cured it.
    New version on sources.
    
    Russ


Reference: /n/sources/patch/sorry/acme-elog
Date: Wed Mar 23 08:04:08 CET 2005
Reviewed-by: rsc

--- /sys/src/cmd/acme/elog.c	Wed Mar 23 08:03:50 2005
+++ /sys/src/cmd/acme/elog.c	Wed Mar 23 08:03:41 2005
@@ -284,6 +284,8 @@
 			}
 			textconstrain(t, b.q0, b.q0+b.nd, &tq0, &tq1);
 			textdelete(t, tq0, tq1, TRUE);
+			q0 = t->q0;
+			q1 = t->q1;
 			break;
 
 		case Insert: