control-C can make equal polynomials give different cyclotomic parts

[dsm054]

In Sage 8.4 on ubuntu, in some cases, two equal polynomials -- one returned by a cached function and the one returned by a direct call -- give different results when you call .cyclotomic_part() on them. Unfortunately it's nondeterministic, and so it could have nothing to do with the function itself. I think it happens after an ill-fated interrupt.

Starting from

c = cached_function(lambda p,q: cyclotomic_polynomial(p).subs(cyclotomic_polynomial(q)))

I eventually got into a state where

sage: broken = cc(3, 935)
sage: assert broken is cc.cache[((3, 935), ())]
sage: working = cyclotomic_polynomial(3).subs(cyclotomic_polynomial(935))
sage: type(broken), type(working)
(<type 'sage.rings.polynomial.polynomial_integer_dense_flint.Polynomial_integer_dense_flint'>,
 <type 'sage.rings.polynomial.polynomial_integer_dense_flint.Polynomial_integer_dense_flint'>)
sage: broken == working, broken - working, broken / working, broken % working
(True, 0, 1, 0)
sage: broken.has_cyclotomic_factor()
True
sage: broken.cyclotomic_part()
1
sage: working.has_cyclotomic_factor()
True
sage: working.cyclotomic_part()
x^16 + x^8 + 1

I've seen this happen several times now-- earlier, in a different session, it happened at (3, 1103). If I use dumps to save the objects and then load them into a separate session, the problem doesn't carry over.

So since it's clearly stateful in some way, I wondered if it has to do with what happens if you use control-C to interrupt a process at an inopportune moment, and indeed, if I run

cc = cached_function(lambda p,q: cyclotomic_polynomial(p).subs(cyclotomic_polynomial(q)))

to create a new function, and then run the following repeatedly, hitting control-C each time:

for i in range(1, 2000):
    print(i)
    assert cc(3, i).has_cyclotomic_factor() == (cc(3, i).cyclotomic_part() != 1)

I eventually see something like

87
88
89
---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
<ipython-input-5-40435234e9d7> in <module>()
      1 for i in range(Integer(1), Integer(2000)):
      2     print(i)
----> 3     assert cc(Integer(3), i).has_cyclotomic_factor() == (cc(Integer(3), i).cyclotomic_part() != Integer(1))
      4 

AssertionError: 
sage: i
89
sage: cc(3, i) == cyclotomic_polynomial(3).subs(cyclotomic_polynomial(i))
True
sage: cc(3, i).cyclotomic_part()
1
sage: cyclotomic_polynomial(3).subs(cyclotomic_polynomial(i)).cyclotomic_part()
x^2 - x + 1

This is suboptimal. :-)

(I wasn't sure what component to use: please edit to whatever makes sense.)

Component: misc

Issue created by migration from https://trac.sagemath.org/ticket/26788

[dsm054]

comment:1

There are several possibly related tickets involving alarms, e.g. #24986, which aren't encouraging.

Unfortunately when I was trying to automate the tests I hit

sage: alarm(3)
sage:                                                                                                          

AlarmInterrupt escaped interact()

sage: 
sage:                                                                                                          
^[[26;1R
**********************************************************************

Oops, Sage crashed. We do our best to make it stable, but...