use weakref for PolyBoRi

[malb]

This patch makes sure only one BooleanPolynomialRing per parameter set is created by returning a reference to a prior created object if there is such a reference.

CC: @burcin @sagetrac-PolyBoRi

Component: commutative algebra

Keywords: PolyBoRi, editor_malb

Issue created by migration from https://trac.sagemath.org/ticket/3284

[sagetrac-PolyBoRi]

comment:2

Hello again,
"Operands come from different manager." is an error message from the deep of the BDD-package behind the PolyBoRi data structures. It means, that one tries to do a binary operation with elements from two different rings. But operation does not necessary mean the y*z operation, it maybe triggered by something else in N() oder _coerce_

Best regards,
Alexander

[malb]

comment:3

My guess is that we somewhere forget to set the global ring maybe?

[malb]

Attachment: pbori_weakref.patch.gz

this patch is supposed to work

[malb]

Changed keywords from PolyBoRi, segfault to PolyBoRi

[malb]

comment:5

Please review the updated patch which fixes the doctest failure.

[burcin]

comment:6

I don't see why the changes to pbori.pyx other than the addition of the R._pbring.activate() at line 4131 are necessary.

In BooleanPolynomialRing_constructor, the if statement (including the elif) at line 430 seems to be redundant, since normalize_names already returns a tuple. It could be that I'm reading the source of normalize_names wrong though.

[malb]

comment:7

True, it is unrelated to this particular weakref patch. I mixed up two things. The renaming only cleans up since vars is a built-in identifier and it is considered bad practice to use it like we used to use it.

[craigcitro]

Changed keywords from PolyBoRi to PolyBoRi, editor_malb

[burcin]

BooleanPolynomialRing user friendly names

[burcin]

comment:9

Attachment: trac3284_BooleanPolynomialRing_normalize_names.patch.gz

BooleanPolynomialRing_constructor in malb's original patch fails when only names are provided, attachment: trac3284_BooleanPolynomialRing_normalize_names.patch fixes this case.

I give malb's patch (followed by mine) a positive review. Someone should review my patch, especially the change to normalize_names.

[malb]

comment:10

Burcin's patch looks good and passes doctests.

[sagetrac-mabshoff]

comment:11

All doctests pass with the patch applied and valgrind gives pbori.pyx a clean bill of health.

Cheers,

Michael

[sagetrac-mabshoff]

comment:12

Merged in Sage 3.0.4.alpha1