search

rtoy / maxima

A Clone of Maxima's repo
Other
0 stars 0 forks source link

cmucl 21e intermittent lisp error in limit #4058

Closed rtoy closed 4 months ago

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:05 Created by dgildea on 2023-10-09 15:05:16 Original: https://sourceforge.net/p/maxima/bugs/4195

I get the following nondeterministic error with cmucl 21e

Maxima branch_5_47_base_432_gd10d5ac7a https://maxima.sourceforge.io
using Lisp CMU Common Lisp 21e (21E Unicode)
Distributed under the GNU Public License. See the file COPYING.
Dedicated to the memory of William Schelter.
The function bug_report() provides bug reporting information.
(%i1) limit(gamma(x)/gamma(1/x), x, 0, plus);
(%o1)                                  0
(%i2) limit(gamma(x)/gamma(1/x), x, 0, plus);
(%o2)                                  0
(%i3) limit(gamma(x)/gamma(1/x), x, 0, plus);
(%o3)                                  0
(%i4) limit(gamma(x)/gamma(1/x), x, 0, plus);

Maxima encountered a Lisp error:

 Type-error in KERNEL::OBJECT-NOT-LIST-ERROR-HANDLER:
    59854938 is not of type LIST

Automatically continuing.
To enable the Lisp debugger set *debugger-hook* to nil.

I do not get the error with ecl or gcl.

The problem seems to start with this commit:

commit 24a57b649a78077ad0fe76a7d849a9bfd1bed574
Author: Barton Willis <58446958+barton-willis@users.noreply.github.com>
Date:   Fri Jul 14 06:10:49 2023 -0500

    Fixes for rtest_limit_extra bugs 52, 64, 67, 234, 277, 278, and 279.
rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:07 Created by robert_dodier on 2023-10-10 02:55:48 Original: https://sourceforge.net/p/maxima/bugs/4195/#a124

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:11 Created by willisbl on 2023-10-11 12:17:55 Original: https://sourceforge.net/p/maxima/bugs/4195/#6f7a

I'm unable to reproduce this with either Clozure CL (version 1.12.2, Windows) or SBLC (2.2.0, Windows). And I don't have a machine that can run CMU CL.

I'm out of ideas. Suggestions on how to sort this?

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:14 Created by rtoy on 2023-10-12 14:05:09 Original: https://sourceforge.net/p/maxima/bugs/4195/#f0d9

I can reproduced this but I'm using snapshot-2023-08, which is slightly newer than 21e. I get something deterministic:

(%i1) limit(gamma(x)/gamma(1/x), x, 0, plus);
Evaluation took 0.1000 seconds (0.1000 elapsed) using 12.052 MB.
(%o1)                                  0
(%i2) limit(gamma(x)/gamma(1/x), x, 0, plus);

Type-error in KERNEL::OBJECT-NOT-LIST-ERROR-HANDLER:  0 is not of type LIST
   [Condition of type TYPE-ERROR]

Restarts:
  0: [MACSYMA-QUIT] Maxima top-level

Debug  (type H for help)

(SPECREPP (0 . 0))
Source: 
; File: /home/toy/src/sourceforge/maxima/src/simp.lisp
(CAAR E)
0]

The backtrace is:

0] back 9

0: (SPECREPP (0 . 0))
1: (SUBST1 ((MEXPT SIMP) $X -1) $Z (0 . 0))
2: (SUBST1 ((MEXPT SIMP) $X -1) $Z ((MEXPT SIMP) $%PI (0 . 0)))
3: (SUBST1 ((MEXPT SIMP) $X -1)
           $Z
           ((MTIMES SIMP) ((MEXPT SIMP) 2 ((RAT SIMP) 1 2))
            ((MEXPT SIMP) $%PI (0 . 0))
            ((MEXPT SIMP) $Z ((MPLUS SIMP) ((RAT SIMP) -1 2) $Z))
            ((MEXPT SIMP) $%E ((MTIMES SIMP) -1 $Z))))
4: (MAXIMA-SUBSTITUTE ((MEXPT SIMP) $X -1)
                      $Z
                      ((MTIMES SIMP) ((MEXPT SIMP) 2 ((RAT SIMP) 1 2))
                       ((MEXPT SIMP) $%PI (0 . 0))
                       ((MEXPT SIMP) $Z ((MPLUS SIMP) ((RAT SIMP) -1 2) $Z))
                       ((MEXPT SIMP) $%E ((MTIMES SIMP) -1 $Z))))
5: (STIRLING0 ((MFACTORIAL SIMP) ((MPLUS SIMP) -1 ((MEXPT SIMP) $X -1))))
6: (STIRLING0
    ((MEXPT SIMP) ((MFACTORIAL SIMP) ((MPLUS SIMP) -1 ((MEXPT SIMP) $X -1)))
     -1))
7: (STIRLING0
    ((MTIMES SIMP)
     ((MEXPT SIMP) ((MFACTORIAL SIMP) ((MPLUS SIMP) -1 ((MEXPT SIMP) $X -1)))
      -1)
     ((MFACTORIAL SIMP) ((MPLUS SIMP) -1 $X))))
8: (TAYLIM
    ((MTIMES SIMP)
     ((MEXPT SIMP) ((MFACTORIAL SIMP) ((MPLUS SIMP) -1 ((MEXPT SIMP) $X -1)))
      -1)
     ((MFACTORIAL SIMP) ((MPLUS SIMP) -1 $X)))
    $X
    $ZEROA
    #<unused-arg>)

Perhaps the backtrace will have some hints onw what's happening. I'll take a closer look later,

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:18 Created by willisbl on 2023-10-12 14:10:48 Original: https://sourceforge.net/p/maxima/bugs/4195/#5145

Maybe it's obvious, but I'd focus on the functions stirling0 and taylim as suspects.

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:21 Created by rtoy on 2023-10-22 23:11:59 Original: https://sourceforge.net/p/maxima/bugs/4195/#5be8

It's not just 21e. I get the same error with 21d (released in 2018-12). Now it's hard to tell if the issue is in cmucl or in maxima. I think the next step is to try to do a bisect of maxima using cmucl 21d to see if there's some obvious change in maxima that causes this issue.

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:25 Created by dgildea on 2023-10-22 23:59:30 Original: https://sourceforge.net/p/maxima/bugs/4195/#7417

Using cmucl 21b, the problem also seems to start with the commit mentioned above.

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:29 Created by rtoy on 2023-10-24 21:40:13 Original: https://sourceforge.net/p/maxima/bugs/4195/#a22b

I've found a solution to this problem, but I'm not sure it's quite right. The solution is modifying stirling so that it conses up a new list for the third arg to maxima-substitute. It appears maxima-substitute can modify the third arg, which, in this case destructively modifies the constant list used in stirling.

But I'm not yet sure if that's right; I need to study what maxima-substitute does to the third arg to see if can really modify the third arg. If so, we need to find all uses of maxima-substitute and make sure the third arg is always a fresh list and not a constant.

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:32 Created by rtoy on 2023-10-25 14:40:08 Original: https://sourceforge.net/p/maxima/bugs/4195/#0a44/0f20

That's similar to what I did, but I just generated the third arg for maxima-substitute anew each time.

As an experiment, I reverted the changes to stirling0. This fixes the issues with cmucl. But now maxima fails the one test: limit(gamma(1/x)/gamma(x),x,0,'plus) which returns minf instead of inf.

I think your proposed version of stirling is much easier to understand than the current version and is worth doing no matter what.

I'm going to take a closer look to see what's special about the changes to stirling0 that causes the bad behavior. It still troubles me that cmucl complains about ignoring the special var e in simplimit%limit. I can't find where e was declared to be special (when building maxima from scratch). Cmucl's debugger actually says it's an undefined variable with a specific value that looks like the third arg of maxima-substitute, but modified from the actual constant list in the source.

rtoy commented 4 months ago

Imported from SourceForge on 2024-07-09 11:12:36 Created by rtoy on 2023-10-25 14:44:52 Original: https://sourceforge.net/p/maxima/bugs/4195/#0a44/0f20/1553

And it's not just cmucl that complains about ssimplim%limit. Clisp says:

;; Compiling file /home/toy/src/sourceforge/maxima/src/limit.lisp ...
WARNING: in SIMPLIM%LIMIT in lines 2968..2970 : Binding variable E can cause
         side effects despite IGNORE declaration since it is declared SPECIAL.
WARNING: in SIMPLIM%LIMIT in lines 2968..2970 : Binding variable X can cause
         side effects despite IGNORE declaration since it is declared SPECIAL.

Don't know why cmucl doesn't think x is a specvar.

In any case, we really need to figure this out. I'll file a different bug for this issue.