I'm keen to keep Matrix_integer_2x2 as a subclass of Matrix. I'd have said that doing it any other way is likely to cause lots of obscure bugs (since developers will quite naturally assume that Matrix_xxx is a subclass of Matrix for any value of xxx, and have probably already done so in existing code).
David On 11 January 2013 13:14, Timo Kluck <tkl...@gmail.com> wrote: > Op vrijdag 11 januari 2013 13:41:42 UTC+1 schreef David Loeffler het > volgende: >> >> >> > I think the reason for subclassing in this case is code-reuse. But >> that's in itself a dangerous thing to do if we change the class' internal > >> behavior by this much. People changing the code for Matrix_dense shouldn't >> need to deal with issues in a non-conforming subclass -- >> > but they will have to when their changes suddenly cause doctests to >> fail. >> >> The probelm with that is that conforming properly to the interface (...) >> results in slowing down 2x2 matrix multiplication by a factor of about 5. >> > > I understand that. That's why I'm suggesting not subclassing at all. Then > you can have all the speed benefit you want without breaking expectations. > As I said, when you don't conform to the interface, the only reason for > subclassing is code reuse, but that too is a shady thing to do when you > change the internals. > > I have a suggestion for a workaround: create a single Mutability object >> which is constant over each Sage session and is set to immutable, and >> assign this to the _mutability slot in _new_c. This means that we still >> preserve most of the speed advantage of the 2x2 integer matrix class, but >> it still conforms to the specification modulo the fact that it's not >> possible to create a mutable Matrix_integer_2x2. The downside is that we'd >> potentially end up with a situation where some Matrix_integer_2x2 objects >> will have "their own" Mutability objects, and some will have a pointer to a >> single global Mutability object, which is a bit mind-bending; but that >> doesn't matter since we will never change the global 2x2 matrix Mutability >> object anyway. >> >> This may be an adequate solution. I prefer not subclassing, though. How > do you feel about that? > > Timo > > -- > You received this message because you are subscribed to the Google Groups > "sage-devel" group. > To post to this group, send email to sage-devel@googlegroups.com. > To unsubscribe from this group, send email to > sage-devel+unsubscr...@googlegroups.com. > Visit this group at http://groups.google.com/group/sage-devel?hl=en. > > > -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To post to this group, send email to sage-devel@googlegroups.com. To unsubscribe from this group, send email to sage-devel+unsubscr...@googlegroups.com. Visit this group at http://groups.google.com/group/sage-devel?hl=en.
