[Maxima-discuss] gnuplot-send-command and broken pipes

[Leo B. <l_b...@us...>]

If gnuplot crashes during a plot/draw operation, the user must fix the
problem manually. Presumably, gnuplot_restart() is meant to be used
here, but this can cause a hang (at least with sbcl), so in the past I
have just restarted maxima. But recently I had invested some time in a
session and did not want to restart so I looked a bit deeper.

The following patch tries to programmatically handle the issue. Of the 3
lisps I have available (ecl, gcl, sbcl) both ecl and sbcl signal a
stream-error and we can handle that; gcl signals an
internal-simple-error which seems too generic to handle so it is just
passed on. I don't know what other lisps do with a broken pipe.

I realize that this opens a can of worms and exposes the deficiencies in
the CL streams specs and all the headaches of cross-platform support,
but it seems like we should be able to do better for most/many cases.


diff --git a/src/plot.lisp b/src/plot.lisp
index fe47a3f3f..ba09b30fd 100644
--- a/src/plot.lisp
+++ b/src/plot.lisp
@@ -148,9 +148,18 @@ sin(y)*(10.0+6*cos(x)),
 (defun send-gnuplot-command (command)
   (if (null *gnuplot-stream*)
       (start-gnuplot-process $gnuplot_command))
-  (when (not (null command))
-    (format *gnuplot-stream* "~a ~%" command)
-    (force-output *gnuplot-stream*)))
+  (handler-case (unless (null command)
+		  (format *gnuplot-stream* "~a ~%" command)
+		  (force-output *gnuplot-stream*))
+    (stream-error (e)
+      ;; allow gnuplot to restart if stream-error
+      ;; ok: sbcl, ecl
+      (warn "~a~%Trying new stream.~%" e)
+      (setq *gnuplot-stream* nil)
+      (send-gnuplot-command command))
+    (error (e)
+      ;; pass along error otherwise
+      (error e))))



-- 
Leo Butler                <l_b...@us...>
SDF Public Access UNIX System -   http://sdf.lonestar.org