segfault in numerical_integral()

[rws]

sage: numerical_integral(sin(x^2)/(x^2), 1, infinity, max_points=10^10)
---------------------------------------------------------------------------
SignalError                               Traceback (most recent call last)
<ipython-input-7-a164db8ca706> in <module>()
----> 1 numerical_integral(sin(x**Integer(2))/(x**Integer(2)), Integer(1), infinity, max_points=Integer(10)**Integer(10))

/home/ralf/sage/local/lib/python2.7/site-packages/sage/gsl/integration.so in sage.gsl.integration.numerical_integral (build/cythonized/sage/gsl/integration.c:2626)()

/home/ralf/sage/local/lib/python2.7/site-packages/sage/ext/c_lib.so in sage.ext.c_lib.sig_raise_exception (build/cythonized/sage/ext/c_lib.c:1208)()

SignalError: Segmentation fault

Better error messages would be needed here, as well.

The segfault happens on OpenSuSE 12.3 (both 6.2 and 6.3) and SMC (Ubuntu) but not on Mac OS X 10.7

[rws]

Segfaults are always critical.

[kcrisman]

What platform is this on? On Mac OS X 10.7 I get (0.285725372660691, 7.519910599701474e-05) for both Sage 5.12 and 6.2.

[kcrisman]

What platform is this on? On Mac OS X 10.7 I get (0.285725372660691, 7.519910599701474e-05) for both Sage 5.12 and 6.2.

This could be a good situation to ask on sage-devel; there are a wide variety of systems there and if everyone tried quickly you might get a lot of feedback. It's too bad the traceback is so short, that's not a lot of info to go on :( as usually I like tracking these things down.

On SMC, which is Ubuntu:

$ sage
┌────────────────────────────────────────────────────────────────────┐
│ Sage Version 6.3, Release Date: 2014-08-10                         │
│ Type "notebook()" for the browser-based notebook interface.        │
│ Type "help()" for help.                                            │
└────────────────────────────────────────────────────────────────────┘
sage: numerical_integral(sin(x^2)/(x^2), 1, infinity, max_points=10^10)
---------------------------------------------------------------------------
SignalError                               Traceback (most recent call last)
<ipython-input-1-a164db8ca706> in <module>()
----> 1 numerical_integral(sin(x**Integer(2))/(x**Integer(2)), Integer(1), infinity, max_points=Integer(10)**Integer(10))
 
/usr/local/sage/sage-6.3.beta6/local/lib/python2.7/site-packages/sage/gsl/integration.so in sage.gsl.integration.numerical_integral (build/cythonized/sage/gsl/integration.c:2626)
()
 
/usr/local/sage/sage-6.3.beta6/local/lib/python2.7/site-packages/sage/ext/c_lib.so in sage.ext.c_lib.sig_raise_exception (build/cythonized/sage/ext/c_lib.c:1208)()
 
SignalError: Segmentation fault
sage: 
Exiting Sage (CPU time 0m0.30s, Wall time 0m7.00s).
~$ uname -a
Linux compute20dc0 3.13.0-32-generic #57-Ubuntu SMP Tue Jul 15 03:51:08 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux

Does yours only happen with Sage 6.3? I don't have that yet locally, though on 6.3.beta5 all is still well.

[rws]

It still happens in 6.4rc1. Changing a,b does not matter, so all functions gsl_integration_qag* are affected but not the qng method:

sage: integrate(sin(x^2)/(x^2), 1, infinity).n()
0.285736646322853
sage: numerical_integral(sin(x^2)/(x^2), 1,infinity, max_points=10^10, algorithm='qng')
(nan, nan)

[rws]

OK, the reason is, gsl_integration_workspace_alloc returns NULL which is then fed into the qag integration. So, the code needs to catch it and give an error message.

[rws]

New commits:

​af3158a

16788: catch gsl error and tell the user about it

[jdemeyer]

should be MemoryError.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​c2bdcba

16788: fix exception type

[jdemeyer]

Needs testing on 32-bit.

[chapoton]

let us try again..

---

New commits:

​87332f3

trac 16788 catch a memory allocation error

[chapoton]

ping ?

[vbraun]

On OSX:

**********************************************************************
File "src/sage/calculus/integration.pyx", line 237, in sage.calculus.integration.?
Failed example:
    numerical_integral(sin(x^2)/(x^2), 1,infinity, max_points=10^10)
Expected:
    Traceback (most recent call last):
    ...
    MemoryError: could not allocate workspace: max_points too big
Got:
    (0.285725372660691, 7.519910599701474e-05)

[vbraun]

Fails on 32-bit, too

[rws]

Well 10^10 > 2^32 so it may have been an int size problem from the start...

[vbraun]

The OSX box is 64-bit

[vdelecroix]

update milestone 8.3 -> 8.4