ECL crashes in integration

[nasser1]

Computing the following integral using "maxima" for integration
causes a Sage core dump.

>sage
┌────────────────────────────────────────────────────────────────────┐
│ SageMath version 8.3.beta6, Release Date: 2018-06-17               │
│ Type "notebook()" for the browser-based notebook interface.        │
│ Type "help()" for help.                                            │
└────────────────────────────────────────────────────────────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Warning: this is a prerelease version, and it may be unstable.     ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
sage: var('x m a b')
(x, m, a, b)
sage: integrate(x^m/sqrt(a + b*x^(2 - m)), x)

;;;
;;; Detected access to protected memory, also kwown as 'bus or segmentation fault'.
;;; Jumping to the outermost toplevel prompt
;;;

When using other algorithms, it works, i.e. Sage does not crash.

sage: integrate(x^m/sqrt(a + b*x^(2 - m)),x,algorithm="giac")
integrate(x^m/sqrt(b*x^(-m + 2) + a), x)

Observed under Linux Manjaro, 17.1 on an intel PC.

I would add the core dump file itself, but I do know if one is created
or where it is saved. Do not see one in my current folder.

Any other information needed, please let me know.

Note: the same error occurs using Sage 8.2-6 (from the
Arch Linux packaging) under Linux Manjaro MATE 17.1.

$ sage
┌────────────────────────────────────────────────────────────────────┐
│ SageMath version 8.2, Release Date: 2018-05-05                     │
│ Type "notebook()" for the browser-based notebook interface.        │
│ Type "help()" for help.                                            │
└────────────────────────────────────────────────────────────────────┘

Note:

• Upstream (Maxima) bug report: https://sourceforge.net/p/maxima/bugs/3442/

Upstream: Fixed upstream, but not in a stable release.

Component: symbolics

Keywords: integrate, segfault

Issue created by migration from https://trac.sagemath.org/ticket/25636

[kcrisman]

comment:1

See also this ask.sagemath post from same author.

I can confirm this in 8.3.beta6 on Mac OS as well.

[rwst]

Valgrind core dump

[rwst]

comment:2

Attachment: t.gz

Manual input in Sage's Maxima does not show this (just unsolved returned).

BTW, the impact of the Rubi test suite is much appreciated.

[rwst]

comment:4

BTW, the variables a,b,c,d,e,... in the Rubi test suite are intended to be real, I think. And the vars m,n,... are integer. In fact, this does not crash:

sage: var('a b', domain='real'); var('m,n', domain='integer')
(a, b)
(m, n)
sage: integrate(x^m/sqrt(a + b*x^(2 - m)),x)
integrate(x^m/sqrt(b*x^(-m + 2) + a), x)

[nasser1]

comment:5

"BTW, the variables a,b,c,d,e,... in the Rubi test suite are intended to be real, I think. "

Hello;

Actually Rubi is not even meant to be used for floating point numbers. Only exact numbers. Please see this

"Rubi is only designed to handle rational, not floating point, numbers. If you substitute 3/10 instead of 0.3 for a, it returns the correct antiderivative without the GCD error. "

The integrate command should have worked as is without assumption made on variables being real. Adding assumptions have no effect on Rubi's integrate result, since it does not use assumptions.

In[19]:= $Assumptions = x == 0;

(*Rubi*)
Int[1/x, x]
Out[20]= Log[x]

(*Mathematica*)
In[21]:= Integrate[1/x, x]
Out[21]= ComplexInfinity

btw, In Mathematica 11.3 and Rubi also, the result for the original integral is

In[11]:= Integrate[x^m/Sqrt[a + b*x^(2 - m)], x]

Out[11]= (x^(1 + m)*Sqrt[1 + (b*x^(2 - m))/a]*
Hypergeometric2F1[1/2, (1 + m)/(2 - m), -(3/(-2 + m)), -((b*x^(2 - m))/a)])/((1 + m)
*Sqrt[a + b*x^(2 - m)])

No assumptions are used. But good to know that with Maxima using assumptions, now it does not crash sage.

Thank you

--Nasser

[nbruin]

comment:6

Of course it fails in maxima because, well ... from sage it fails in maxima. There's a theoretical possibility that sage corrupts maxima's memory, but we've never observed such bugs, so that's very unlikely. In this case, with

besselexpand : true;
display2d : false;
domain : complex;
keepfloat : true;
load(to_poly_solve);
load(simplify_sum);
load(abs_integrate);
load(diag);

(That's most of the code sage feeds to maxima for initialization, see github) we get

(%i9) integrate(x^m/sqrt(a + b*x^(2 - m)),x);
...
Segmentation fault (core dumped)

For

Maxima 5.41.0 http://maxima.sourceforge.net
using Lisp SBCL 1.4.2-1.fc27

this fails just as badly.

For all these integrals, the right place to report them is really maxima. You'll just have to dig a little to find the combinations of settings that trigger the error in maxima. Of course, you might get the reply back "using these options together doesn't make sense, don't do that" (although I don't think Robert has ever said that), in which case sage might have to reconsider its use of maxima.

[nasser1]

comment:7

"For all these integrals, the right place to report them is really maxima."

Hello;

But when I try the integral directly into Maxima 5.41.0, with lisp ECL 16.1.2, on the same Linux box (Manjora 17.1), Maxima does not crash.

It hanged for long time thinking, so I ended terminating the command eventually. But it did not crash or even give an error.

Only when called from SageMath, there is a crash and it happens very quickly after the call.

May be SageMath is using different Maxima version. I do not know. I do not know how Sage communicate with Maxima, or what version of Maxima sage uses. I assumed it is the same as the Maxima installed on same Linux box, or may be Sage is just linked to a Maxima .so lib

If I report this problem to Maxima, they could tell me that it does not crash in Maxima, which is true, and to report it to SageMath.

Thank you for looking at this.

[mantepse]

comment:8

@Nasser, I would like to ask you a favour: please get back to the tickets for the bugs you reported, once they are fixed. The bugfixes won't make it into sage mainline otherwise (or only after a long time). It's enough to try the bugfix (eg., using git trac try followed by the ticket number, see http://doc.sagemath.org/html/en/developer/git_trac.html), and if it works for you, mention this as a comment on the ticket. That would be extremely helpful!