On Mon, Feb 4, 2013 at 10:24 PM, Nils Bruin <nbr...@sfu.ca> wrote:
> On Feb 4, 10:01 pm, Simon King <simon.k...@uni-jena.de> wrote:
>
>> Note that in some cases the _element_constructor of a complicated ring
>> would try to convert the input into ZZ before converting it into self.
>
> Ouch. by generic conversion?
>
>> So, in order to strictly avoid it, we would need to carefully clean all
>> of Sage. That's why I tried to suggest a mechanism that would allow to
>> deallocate a conversion even when a conversion/coercion into ZZ is
>> cached.
>
> I guess all we require is that phi.domain() returns the domain, so if
> internally this is stored via a weakref, there shouldn't be an issue.
> If we just throw RuntimeError("Domain has disappeared") if it ends up
> not being there anymore, I doubt we'd ever have problems. That way at
> least we won't be keeping domains alive unless there's a reason for it
> inside the map. I doubt any code is regularly inspecting domains of
> coercion maps, so performance loss should be minimal.
>
> I'd say: encourage maps to store their domains via weakref, but not
> require it. Oddly enough, this means we should advise that maps should
> NOT typecheck that their argument lies in the domain and also not try
> to coerce/convert into the domain before applying the map, because
> that would require accessing the stored domain. Is that a problem?
> does that mean that coercion maps that are stored internally need to
> be more bare bones than maps offered for public consumption?Currently the cdef _call_ assumes the argument is in the domain (sometimes giving incorrect results or segfaulting if it's not, for performance reasons), whereas __call__ coerces its argument into the domain if it is not already there. I think this behavior is worth preserving, so something like phi(2), phi: QQ -> R would still work. Also, not being able to get the domain of a morphism just seems really bad (unless perhaps it's a special light-weight coercion morphism, but even then). Wouldn't it be sufficient if the morphism itself was weekly referenced? >> > (R,S,+) -> c >> >> I thought we are doing this already? Certainly TripleDict is used with a >> binop as a third argument, in some contexts. > > Yes, but I was proposing other semantics for the value stored in it, > to avoid having strong references to morphisms outside their > codomains. Essentially, the _coercion_maps cache would map to special values interpreted by the coercion_maps() method, right? I suppose if we're going for speed we could store this in the (renamed?) action TripleDict if the domain is a Parent and the operation has a special single-underscore method. - Robert -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To post to this group, send email to sage-devel@googlegroups.com. Visit this group at http://groups.google.com/group/sage-devel?hl=en. For more options, visit https://groups.google.com/groups/opt_out.
