bug in pickling quotient module over pid

[williamstein]

This is from the "report a problem" link in the notebook:

If you have a vector space, that is a quotient of a subspace of
another vector space, then after coercing elements into it, something
goes wrong in (un)pickling it.

sage: V = VectorSpace(QQ, 2)
sage: W = V.subspace([V([1,1])])
sage: Z = W.subspace([])
sage: WmodZ = W / Z
sage: WmodZ(W(0))
(0)
sage: loads(dumps(WmodZ))
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)

/home/bosman/sage/<ipython console> in <module>()

/home/bosman/sage-4.2-linux-Ubuntu_9.04-i686-Linux/local/lib/python2.6/site-packages/sage/structure/sage_object.so
in sage.structure.sage_object.loads
(sage/structure/sage_object.c:8769)()

/home/bosman/sage-4.2-linux-Ubuntu_9.04-i686-Linux/local/lib/python2.6/site-packages/sage/modules/free_module.pyc
in __hash__(self)
  4576             True
  4577         """
-> 4578         return hash(self.__basis)
  4579
  4580     def construction(self):

AttributeError: 'FreeModule_submodule_field' object has no attribute
'_FreeModule_submodule_with_basis_pid__basis'

Component: linear algebra

Author: Simon Brandhorst

Branch/Commit: ea6a00b

Reviewer: Travis Scrimshaw

Issue created by migration from https://trac.sagemath.org/ticket/7516

[williamstein]

comment:1

NOTE: This bug does not happen for Free modules over ZZ. It's only over a field where the issue happens.

sage: V = FreeModule(ZZ, 2)
sage: W = V.submodule([V([1,1])])
sage: Z = W.submodule([])
sage: WmodZ = W / Z
sage: loads(dumps(WmodZ))
Finitely generated module V/W over Integer Ring with invariants (0)
sage: WmodZ(W(0))
(0)
sage: loads(dumps(WmodZ))
Finitely generated module V/W over Integer Ring with invariants (0)

[williamstein]

Attachment: trac_7516.patch.gz

[simon-king-jena]

comment:3

If I understand the patch correctly, it violates a very essential assumption about hash codes: The hash has to be the same for objects that are equal. But with the patch, it may be possible that during unpickling the hash "0" is returned, while afterwards a completely different value is returned for the same object.

This is giving me headache. Isn't it be possible that things are put in the wrong hash bucket?

Wouldn't it be a cleaner solution to ensure that self.!__basis is defined during unpickling before the hash is requested?

Could you explain why your solution is correct?

[simonbrandhorst]

comment:8

sage: V = VectorSpace(QQ, 2)
sage: W = V.subspace([V([1,1])])
sage: Z = W.subspace([])
sage: WmodZ = W / Z
sage: WmodZ(W(0))
(0)
sage: loads(dumps(WmodZ))

Vector space quotient V/W of dimension 1 over Rational Field where
V: Vector space of degree 2 and dimension 1 over Rational Field
Basis matrix:
[1 1]
W: Vector space of degree 2 and dimension 0 over Rational Field
Basis matrix:
[]

Works for me. Add a doctest and close?

[tscrim]

comment:9

Replying to @simonbrandhorst:

Works for me. Add a doctest and close?

Yep. Addendum - Also works for me.

[simonbrandhorst]

Branch: u/sbrandhorst/bug_in_pickling_quotient_module_over_pid

[simonbrandhorst]

New commits:

ea6a00b

Added a doctest for pickling.

[simonbrandhorst]

Commit: ea6a00b

[tscrim]

comment:12

Once you put your real name as author, you can set a positive review.

[tscrim]

Reviewer: Travis Scrimshaw

[simonbrandhorst]

Author: Simon Brandhorst

[vbraun]

Changed branch from u/sbrandhorst/bug_in_pickling_quotient_module_over_pid to ea6a00b