Factoring bug in Pari stack

[David Roe]

I've run across a strange bug that seems to be hardware dependent: it is repeatable on an Ubuntu server, but I can't replicate it on my Macbook Pro or on Cocalc.  It occurs when factoring rational polynomials, and gives an error message from pari: "bug in gerepile, significant pointers lost, please report" (full traceback and input below).  To trigger the bug, one factors two different polynomials consecutively (factoring the second on its own does not trigger the problem).

The bug seems to be quite sensitive to the polynomials involved.  Deleting any subset of factors seems to not trigger the bug, and multiplying through by the denominator causes it to disappear as well (though other examples still trigger the bug even when all coefficients are integral).

When I try factoring in gp, I don't get an error (thus why I'm writing to sage-devel rather than Pari).

If anyone has ideas on how to diagnose this, or reproduce it in a format more easily reportable to the Pari developers, I'd be happy to hear it.  In the meantime, calling squarefree_decomposition and then factoring each of the resulting terms seems to be a functional workaround for my purposes.

David

System info:

Freshly built Sage 8.7

Ubuntu 18.04.1 LTS (GNU/Linux 4.15.0-30-generic x86_64)

CPUs: 48 Intel Xeon CPU E5-2697 v2 @ 2.70GHz

RAM: 256GB in 16 Hynix DDR3 DIMMs (the bug occurs when only a small amount of memory is in use)

Hard drive: Samsung SSD 840 EVO 1TB (381GB free)

Input and traceback:

sage: R.<x> = QQ[]

sage: f = (x - 1)^8 * (x^2 - 2/3*x + 1)^2 * (x^2 + 2/3*x + 1)^2 * (x^4 + 3*x^3 + 37/9*x^2 + 3*x + 1) * (x^4 + 1/3*x^3 - x^2 + 1/3*x + 1)^2 * (x^16 + 2*x^14 + 211/81*x^12 + 28/9*x^10 + 269/81*x^8 + 28/9*x^6 + 211/81*x^4 + 2*x^2 + 1)^2

sage: g = (x - 1)^32 * (x^2 + 5/3*x + 1)^16

sage: factor(f)

(x - 1)^8 * (x^2 - 2/3*x + 1)^2 * (x^2 + 2/3*x + 1)^2 * (x^4 + 3*x^3 + 37/9*x^2 + 3*x + 1) * (x^4 + 1/3*x^3 - x^2 + 1/3*x + 1)^2 * (x^16 + 2*x^14 + 211/81*x^12 + 28/9*x^10 + 269/81*x^8 + 28/9*x^6 + 211/81*x^4 + 2*x^2 + 1)^2

sage: factor(g)

---------------------------------------------------------------------------

PariError                                 Traceback (most recent call last)

<ipython-input-5-41d64ee9c816> in <module>()

----> 1 factor(g)

/home/sage/sage-8.7/local/lib/python2.7/site-packages/sage/arith/misc.pyc in factor(n, proof, int_, algorithm, verbose, **kwds)

   2563     except TypeError:

   2564         # Maybe the factor() method doesn't have a proof option

-> 2565         return m(**kwds)

   2566 

   2567 

/home/sage/sage-8.7/local/lib/python2.7/site-packages/sage/rings/polynomial/polynomial_element.pyx in sage.rings.polynomial.polynomial_element.Polynomial.factor (build/cythonized/sage/rings/polynomial/polynomial_element.c:35617)()

   4252         R = self._parent.base_ring()

   4253         if hasattr(R, '_factor_univariate_polynomial'):

-> 4254             return R._factor_univariate_polynomial(self, **kwargs)

   4255 

   4256         G = None

/home/sage/sage-8.7/local/lib/python2.7/site-packages/sage/rings/rational_field.pyc in _factor_univariate_polynomial(self, f)

   1348 

   1349         """

-> 1350         G = list(f._pari_with_name().factor())

   1351 

   1352         # normalize the leading coefficients

cypari2/gen.pyx in cypari2.gen.Gen.factor()

cypari2/handle_error.pyx in cypari2.handle_error._pari_err_handle()

PariError: bug in gerepile, significant pointers lost, please report

[Daniel Krenn]

On 06.05.19 10:37, David Roe wrote:
> I've run across a strange bug that seems to be hardware dependent:

I can reproduce the error on both systems that I've just checked:
Intel(R) Core(TM) i7-8550U CPU @ 1.80GHz (Laptop)
Intel(R) Core(TM) i5-4690 CPU @ 3.50GHz (Desktop)
Both run Linux Mint 19. Both with SageMath >=8.7. Could not reproduce it
with 8.6 (on the i5) and with 8.4 (on the i7) (which I still had available).

Best, Daniel

[Dima Pasechnik]

I opened https://trac.sagemath.org/ticket/27778 for this

I can also reproduce this on Gentoo Linux with Sage 8.8.beta3

> --
> You received this message because you are subscribed to the Google Groups "sage-devel" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.
> To post to this group, send email to sage-...@googlegroups.com.
> Visit this group at https://groups.google.com/group/sage-devel.
> For more options, visit https://groups.google.com/d/optout.