22.1.50; calc-math.el patch: fix non-termination

22.1.50; calc-math.el patch: fix non-termination

[Markus Triska]

With the OS I'm currently working on, the result of (expt 10.0 x) is
1.0e+INF for x >= 309, i.e., no range error is triggered. This patch
to calc-math.el prevents non-termination when loading the file.

2007-09-26  Markus Triska  <markus.triska@gmx.at>

	* calc/calc-math.el (math-largest-emacs-expt)
	(math-smallest-emacs-expt): more robust computation

diff --git a/lisp/calc/calc-math.el b/lisp/calc/calc-math.el
index 3e4743d..44ccbcc 100644
--- a/lisp/calc/calc-math.el
+++ b/lisp/calc/calc-math.el
@@ -54,28 +54,32 @@
 
 (defvar math-largest-emacs-expt
   (let ((x 1))
-    (while (condition-case nil
-               (expt 10.0 x)
-             (error nil))
+    (while (and (condition-case nil
+		    (expt 10.0 x)
+		  (error nil))
+		(< (expt 10.0 x) (expt 10.0 (* 2 x))))
       (setq x (* 2 x)))
     (setq x (/ x 2))
-    (while (condition-case nil
-               (expt 10.0 x)
-             (error nil))
+    (while (and (condition-case nil
+		    (expt 10.0 x)
+		  (error nil))
+		(< (expt 10.0 x) (expt 10.0 (* 2 x))))
       (setq x (1+ x)))
     (- x 2))
   "The largest exponent which Calc will convert to an Emacs float.")
 
 (defvar math-smallest-emacs-expt
   (let ((x -1))
-    (while (condition-case nil
-               (expt 10.0 x)
-             (error nil))
+    (while (and (condition-case nil
+		    (expt 10.0 x)
+		  (error nil))
+		(> (expt 10.0 x) 0.0))
       (setq x (* 2 x)))
     (setq x (/ x 2))
-    (while (condition-case nil
-               (expt 10.0 x)
-             (error nil))
+    (while (and (condition-case nil
+		    (expt 10.0 x)
+		  (error nil))
+		(> (expt 10.0 x) 0.0))
       (setq x (1- x)))
     (+ x 2))
     "The smallest exponent which Calc will convert to an Emacs float.")

Re: 22.1.50; calc-math.el patch: fix non-termination

[Jay Belanger]

Markus Triska <markus.triska@gmx.at> writes:

> With the OS I'm currently working on, the result of (expt 10.0 x) is
> 1.0e+INF for x >= 309, i.e., no range error is triggered.

I don't know why I thought `expt' would necessarily give a range error
when it's too large; the documentation is silent on this matter.
I take it there are three possibilities for large x:  expt will give
an error, it will give infinity, or it will give 0.

> This patch to calc-math.el prevents non-termination when loading the
> file.

There was talk about adjusting `expt' to give a range error for large
x; if that doesn't happen then something like Markus's patch (I'll
tweak it to take care of the 0 case, for example) will be needed.

Using the 15 line convention, this patch is borderline tiny, but
Markus's name appears several times in the lisp Changelogs.  Do we
have copyright papers for him/you?

Jay

Re: 22.1.50; calc-math.el patch: fix non-termination

[Richard Stallman]

We have papers from Markus Triska.  If nobody points out any problem
in this patch in 3 days, please install it.

Re: 22.1.50; calc-math.el patch: fix non-termination

[Markus Triska]

Jay Belanger <jay.p.belanger@gmail.com> writes:

> There was talk about adjusting `expt' to give a range error for large
> x; if that doesn't happen then something like Markus's patch (I'll
> tweak it to take care of the 0 case, for example) will be needed.

The 0 case cannot arise for math-largest-emacs-expt, and it is handled
for math-smallest-emacs-expt. I have now tested on another OS and
suggest the following improved patch, which should work in all cases:

2007-09-27  Markus Triska  <markus.triska@gmx.at>

	* calc/calc-math.el (math-largest-emacs-expt)
	(math-smallest-emacs-expt): more robust computation

diff --git a/lisp/calc/calc-math.el b/lisp/calc/calc-math.el
index 3e4743d..1cfc582 100644
--- a/lisp/calc/calc-math.el
+++ b/lisp/calc/calc-math.el
@@ -53,14 +53,17 @@
 ;;; is an Emacs float, for acceptable d.dddd....
 
 (defvar math-largest-emacs-expt
-  (let ((x 1))
+  (let ((x 1)
+        err)
     (while (condition-case nil
-               (expt 10.0 x)
-             (error nil))
+               (< (expt 10.0 x) (expt 10.0 (* 2 x)))
+             (error (setq err t)
+                    nil))
       (setq x (* 2 x)))
-    (setq x (/ x 2))
+    (unless err
+      (setq x (/ x 2)))
     (while (condition-case nil
-               (expt 10.0 x)
+               (< (expt 10.0 x) (expt 10.0 (1+ x)))
              (error nil))
       (setq x (1+ x)))
     (- x 2))
@@ -69,12 +72,12 @@
 (defvar math-smallest-emacs-expt
   (let ((x -1))
     (while (condition-case nil
-               (expt 10.0 x)
+               (> (expt 10.0 x) 0.0)
              (error nil))
       (setq x (* 2 x)))
     (setq x (/ x 2))
     (while (condition-case nil
-               (expt 10.0 x)
+               (> (expt 10.0 x) 0.0)
              (error nil))
       (setq x (1- x)))
     (+ x 2))