[with patch, positive review] MPolynomialRing_libsingular __dealloc__ is buggy, can lead to crash

[cwitty]

In __dealloc__, if currRing is NULL on entry, then currRing will be the ring we just deleted on exit. The patch fixes this bug, so that currRing never points to freed memory.

It took me quite a while to come up with a small reproducible test case for the problem; here it is. (This test case is also in the patch, as a doctest.)

import gc
from sage.rings.polynomial.multi_polynomial_libsingular import MPolynomialRing_libsingular
R1 = MPolynomialRing_libsingular(GF(5), 2, ('x', 'y'), TermOrder('degrevlex', 2))
R2 = MPolynomialRing_libsingular(GF(11), 2, ('x', 'y'), TermOrder('degrevlex', 2))
R3 = MPolynomialRing_libsingular(GF(13), 2, ('x', 'y'), TermOrder('degrevlex', 2))
gc.collect()
foo = R1.gen(0)
del foo
del R1
gc.collect()
del R2
gc.collect()
del R3
gc.collect()

[malb]

Doctests pass, patch reads good.

[mabshoff]

Merged in Sage 3.4.2.alpha0.

Cheers,

Michael

[mabshoff]

Mmmh, I seem to be seeing random doctest failure with 3.4.1 + only this patch merged, so I am reopening this ticket for now. Can someone do extensive testing on sage.math to see if they can reproduce it?

Cheers,

Michael

[mabshoff]

Ok, given that this causes issues for me I am changing it to 'needs work' for now.

Cheers,

Michael

[cwitty]

What sort of errors? (Crashes? Wrong answers? Any particular files?)

[mabshoff]

I have come to believe that the issues I saw were due to other issues, so I am reinstating the positive review. But I will do some more extensive testing before merging this. Sorry for the noise.

Cheers,

Michael

[mabshoff]

Merged in Sage 4.0.alpha0.

Cheers,

Michael