#16275: Hom: introduce a check argument to simplify the unpickling detection
logic
-------------------------------------+-------------------------------------
Reporter: nthiery | Owner:
Type: enhancement | Status: new
Priority: major | Milestone: sage-6.2
Component: categories | Resolution:
Keywords: homset, pickling | Merged in:
Authors: Nicolas M. Thiéry | Reviewers:
Report Upstream: N/A | Work issues:
Branch: | Commit:
u/nthiery/hom__introduce_a_check_argument_to_simplify_the_unpickling_detection_logic|
ae3ca8080a8412891722695d1e24e1dfe374d787
Dependencies: | Stopgaps:
-------------------------------------+-------------------------------------
Comment (by SimonKing):
Replying to [comment:3 tscrim]:
> I agree with these changes, although I don't understand pickling well
enough to answer those issues. Simon, could you render judgement on this
or clarify the issue?
Let's try...
The problem occurs if Python's default way of pickling/copying is used:
Assume that you have an object `O`, pickle it, and want to unpickle it
into a new object `N`.
When pickling `O` in Python's default way, `type(O)` and `O.__dict__` are
stored. When unpickling, `N` is created as a new instance of `type(O)`,
and `N.__dict__` is filled by unpickling the pickle of `O.__dict__`. Of
course, when unpickling `O.__dict__`, `N` takes the role of `O`.
The problem: It could be that you need `N.__dict__['foo']` in order to
unpickle `N.__dict__['bar']`. To be more concrete, it could be that some
morphism `phi` from `O` to somewhere else is stored as an attribute
`O.foo`. Hence, to unpickle `phi`, you also need to unpickle some
`Hom(O,P,C)` for some category `C`. So, `Hom(N,P,C)` will be created. But
at that point, `N.__dict__` is not completely filled. Perhaps `N` can not
tell whether it is an object of `C` before `N.bar` is available. Thus,
some assertion during creation of the homset will fail.
The obvious approach towards a solution: When we have a pickle, then the
assertions need not be made, as we ''know'' that the to-be-unpickled
object belongs to the category of the to-be-unpickled homset.
In #14793, it was suggested to skip the assertion if the category of
domain or codomain is not initialised, since this indicates that we are in
the process of unpickling domain and/or codomain.
Here, it is suggested to be more explicit and have an optional argument
for skipping the assertion.
I think explicit is better than implicit. Moreover, it has another use
case (as demonstrated by Nicolas) and should also be faster than the old
solution. Hence, I am all in favour of the new solution.
The only problem is: (How) Do we care about old pickles whose unpickling
requires to skip the assertion, but which are unaware of the "check=False"
argument?
I suggest the following scheme:
{{{
if check:
if domain not in category:
if domain's category is initialised:
raise error, since that means we are not unpickling an old
pickle
if codomain not in category:
if codomain's category is initialised:
raise error
proceed with creating the homset
}}}
Hence, the assertion is skipped if this is explicitly requested, and
before raising an actual error it is verified if we need to skip the
assertion implicitly.
--
Ticket URL: <http://trac.sagemath.org/ticket/16275#comment:9>
Sage <http://www.sagemath.org>
Sage: Creating a Viable Open Source Alternative to Magma, Maple, Mathematica,
and MATLAB
--
You received this message because you are subscribed to the Google Groups
"sage-trac" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/sage-trac.
For more options, visit https://groups.google.com/d/optout.
