internal side effect in roots?

[Paul Zimmermann]

Consider the following with Sage 4.4.4:

sage: var('a6,a5,a4,x')
(a6, a5, a4, x)
sage: e=15*a6*x^2 + 5*a5*x + a4
sage: e.roots(x)
[(-1/30*(sqrt(-12*a4*a6 + 5*a5^2)*sqrt(5) + 5*a5)/a6, 1), (1/30*(sqrt(-12*a4*a6 + 5*a5^2)*sqrt(5) - 5*a5)/a6, 1)]

This is fine. However:

sage: var('f6,f5,f4,x')
(f6, f5, f4, x)
sage: e=15*f6*x^2 + 5*f5*x + f4
sage: e.roots(x)
[(1/30*(-I*sqrt(35) - 5)/binomial(n, k), 1), (1/30*(I*sqrt(35) - 5)/binomial(n, k), 1)]

WTF???

[Mike Hansen]

I'm pretty sure this is just #8734.

[Paul Zimmermann]

Replying to mhansen:

I'm pretty sure this is just #8734.

I tried the patch coming with #8734, but this does not resolve this issue.

Paul

[Mitesh Patel]

Replying to zimmerma:

Replying to mhansen:

I'm pretty sure this is just #8734.

I tried the patch coming with #8734, but this does not resolve this issue.

Could you try this again? After applying #8734 to 4.5.3.alpha0 (and ignoring the patch rejects), I get

sage: var('f6,f5,f4,x')
(f6, f5, f4, x)
sage: e=15*f6*x^2 + 5*f5*x + f4
sage: e.roots(x)
[(-1/30*(sqrt(-12*f4*f6 + 5*f5^2)*sqrt(5) + 5*f5)/f6, 1), (1/30*(sqrt(-12*f4*f6 + 5*f5^2)*sqrt(5) - 5*f5)/f6, 1)]

[Burcin Erocal]

The patch attached to #8734 indeed fixes this problem. However there are simpler solutions, and I'm not convinced that #8734 is necessary, given that #7377 could solve this in a much cleaner way. Also see the discussion linked from comment:3:ticket:8734:

​http://groups.google.com/group/sage-devel/browse_thread/thread/67f0a63d00b8d835/06557a921a582f87

In particular, Robert Dodier's suggestion to apply this patch to maxima:

--- share/contrib/Zeilberger/testZeilberger.mac 9 Feb 2007 22:32:34
-0000       1.4
+++ share/contrib/Zeilberger/testZeilberger.mac 15 Jan 2010 19:10:53
-0000
@@ -110,3 +110,8 @@

 FULL_TEST : append(GOSPER_TEST,EASY_TEST,
                    HARD_TEST,EXTREME_TEST);
+
+kill (g1, g2, g3, g4, g5, g6, g7,
+    f1, f2, f3, f4, f5, f6, f7, f8, f9, f10,
+    h1, h2, h3, h4, h5, h6, h6b, h7, h8, h9, h10, h11, h12, h13,
+    d1, d2); 

I can confirm that adding these lines to $SAGE_LOCAL/share/maxima/5.20.1/share/contrib/Zeilberger/testZeilberger.mac solves this issue.

Perhaps this patch is already in a more recent version of maxima?

[Paul Zimmermann]

I confirm that with the #8734 patch applied to 4.5.3 (and ignoring the patch reject) it works fine.

However I'm convinced by Burcin's argument. I am ready to review a patch based on this patch applied to Maxima.

Paul

[Burcin Erocal]

Here is a maxima package that patches the file share/contrib/Zeilberger/testZeilberger.mac as suggested by Robert Dodier:

​http://sage.math.washington.edu/home/burcin/maxima-5.20.1.p1.spkg

attachment:trac_9538-maxima_kill.patch​ adds a doctest to check the example given in the description.

[Paul Zimmermann]

I tried sage -i /tmp/maxima-5.20.1.p1.spkg but got:

...
;;; OPTIMIZE levels: Safety=2, Space=0, Speed=3, Debug=2
;;; End of Pass 1.
;;; Note: Creating tag: "_eclFGkv5HJdeKquW_9rDWPWz" for #P"binary-ecl/maxima-package.o"
;;; Internal error: Unable to find include directory
;      - Binary file binary-ecl/maxima-package.fas is old or does not exist. 
;        Compile (and load) source file /usr/local/sage-4.5.3/sage/spkg/build/maxima-5.20.1.p1/src/src/maxima-package.lisp instead? y

;      - Should I bother you if this happens again? y

;      - Compiling source file
;        "/usr/local/sage-4.5.3/sage/spkg/build/maxima-5.20.1.p1/src/src/maxima-package.lisp"
;;; Compiling /usr/local/sage-4.5.3/sage/spkg/build/maxima-5.20.1.p1/src/src/maxima-package.lisp.
;;; OPTIMIZE levels: Safety=2, Space=0, Speed=3, Debug=2
;;; End of Pass 1.
;;; Note: Creating tag: "_eclFGkv5HJdeKquW_IVMWPWz" for #P"binary-ecl/maxima-package.o"
;;; Internal error: Unable to find include directory
;      - Loading binary file "binary-ecl/maxima-package.fas" An error occurred during initialization:
Filesystem error with pathname #P"/usr/local/sage-4.5.3/sage/spkg/build/maxima-5.20.1.p1/src/src/binary-ecl/maxima-package.fas".
Either
 1) the file does not exist, or
 2) we are not allow to access the file, or
 3) the pathname points to a broken symbolic link..
make[1]: *** [binary-ecl/maxima] Error 1
make[1]: Leaving directory `/usr/local/sage-4.5.3/sage/spkg/build/maxima-5.20.1.p1/src/src'
make: *** [all-recursive] Error 1
***********************************************************
Failed to make Maxima.
***********************************************************

Did I something wrong? How to install the patched spkg?

Paul

[Burcin Erocal]

patch for maxima spkg

[Burcin Erocal]

Replying to zimmerma:

Did I something wrong? How to install the patched spkg?

I don't have any experience with the maxima spkg. I just applied the patch in attachment:maxima_package.patch​. It works here, but just complains about not being able to build maxima.fasb:

cp: cannot stat `maxima.fasb': No such file or directory

Perhaps someone more knowledgeable can help out. Building maxima as a library is #8645.

[Paul Zimmermann]

maybe we should wait that #8645 is fixed and merged within Sage to review that ticket.

Paul

[Mitesh Patel]

Adding

kill (g1, g2, g3, g4, g5, g6, g7,
    f1, f2, f3, f4, f5, f6, f7, f8, f9, f10,
    h1, h2, h3, h4, h5, h6, h6b, h7, h8, h9, h10, h11, h12, h13,
    d1, d2);

directly to the bottom of

SAGE_ROOT/local/share/maxima/5.20.1/share/contrib/Zeilberger/testZeilberger.mac

without installing ​http://sage.math.washington.edu/home/burcin/maxima-5.20.1.p1.spkg fixes the problem in the description.

[Mitesh Patel]

Replying to zimmerma:

I tried sage -i /tmp/maxima-5.20.1.p1.spkg but got:

[...]

Did I something wrong? How to install the patched spkg?

I have the same problem with the forthcoming 4.6.alpha2, if I've renamed SAGE_ROOT since that copy of Sage was first compiled (and another installation with the previous name does not exist). In this case, even reinstalling the included Maxima, with

./sage -f spkg/standard/maxima-5.20.1.p0.spkg

triggers the behavior. Paul, did you happen to move or rename your Sage root directory? Does Burcin's p1 package install successfully and fix the roots problem with a Sage that has not been moved?

[Mitesh Patel]

Nils, do you have any thoughts about the problem in comment 7ff?

[Paul Zimmermann]

Paul, did you happen to move or rename your Sage root directory?

no, I did all my experiments in /tmp/sage-4.6.alpha1, where I built sage-4.6.alpha1 from source. I have no SAGE_ROOT environment variable.

Paul

[Paul Zimmermann]

Burcin, do you agree that we try Mitesh's solution from comment 10, which avoids installing a new Maxima spkg? If so, if someone can provide a patch, I will review it.

Paul

[Mitesh Patel]

Replying to zimmerma:

Burcin, do you agree that we try Mitesh's solution from comment 10, which avoids installing a new Maxima spkg? If so, if someone can provide a patch, I will review it.

Except for SAGE_LOCAL/bin, the files under SAGE_LOCAL are not under version control. I was mainly trying to check whether Robert's fix works with an already installed Maxima in Sage.

So far, it seems the build problem is orthogonal to the problem in the description.

[Paul Zimmermann]

So far, it seems the build problem is orthogonal to the problem in the description.

agreed, but how can we proceed in practice so that I can review this ticket?

Paul

[Burcin Erocal]

apply only this patch -- forget about the package :)

[Burcin Erocal]

I uploaded an alternate patch, attachment:trac_9538-maxima_kill.take2.patch​. This issues the kill command Robert Dodier recommended when initializing maxima. There is no need for any changes to the maxima package.

[Paul Zimmermann]

ok for the last patch, which fixes the problem, and all tests pass (tested with Sage 4.4.4).

Paul

PS: a minor remark, is there a mechanism to update the calculus.py patch once the problem is fixed upstream?

[Burcin Erocal]

Thank you for the review.

Replying to zimmerma:

PS: a minor remark, is there a mechanism to update the calculus.py patch once the problem is fixed upstream?

If it is fixed upstream, this doctest should fail:

sage: maxima = Maxima(init_code = ['load(simplify_sum)']) 
sage: maxima('f1') 
binomial(n,k) 

Then we can remove the line with the kill command.

[Paul Zimmermann]

If it is fixed upstream, this doctest should fail: [...]

excellent!

Paul

[Volker Braun]

For the record, this fails with maxima-5.22.1:

File "/home/vbraun/opt/sage-4.6.rc0/devel/sage-main/sage/calculus/calculus.py", line 368:
    sage: maxima('f1')
Expected:
    binomial(n,k)
Got:
    f1

None of the patches from maxima_package.patch are in the updated maxima-5.22.1 package. See #10187 for the updated spgk. I take it that this is the expected behaviour and no further patches to maxima are necessary.