On Saturday, 7 August 2021 at 01:58:42 UTC-7 Markus Wageringel wrote: > Generally, I would be in favor of having an option for immutable vectors > by default, as I had similar uses before. Usually, I just convert > everything to tuples. > > However, I am not sure if having an additional flag on the parent is worth > the extra burden on maintenance. Already the sparsity flag is not always > handled correctly (#29359, #29360), immutability may be even easier to miss. >
It's also less important. As long as the usual arithmetic within the parent respects it, it's fine. > In my own code, I tend to construct vectors by `vector(ring, [..])`, which > relies on the uniqueness of the parent and will not preserve immutability > of vectors. Though, it is fair to say that this code is already broken as > it does not preserve sparsity either. > And it's inefficient if you have the parent lying around anyway! The globar constructor has to mangle the construction parameters, use them as a key to look up in the global cache, and then get you the parent. If you have the desired parent already in hand, none of that is necessary! So I'd say that's just a bad habit. Whether "vector" should by default choose a vector space with default mutable or immutable vectors is another matter (it's safer to have them immutable unless the user really wants them mutable, so that would be the sensible default, but it would go against legacy, so may be harder). Another point is coercion: if we have V = VectorSpace(GF(5),3,immutable_vectors = True) W = VectorSpace(GF(5),3,immutable_vectors = False) what should V.gen(1)+W.gen(1) be? Does V get priority? Otherwise, I think having the flag will be a life-saver in various interactive scenarios and low-impact outside. So I think it should be fairly uncontroversial to implement it. Regarding the performance discussion, (although this was not the question) using vectors also has the potential of being much slower than using numpy arrays directly. > > > sage: a = vector(CDF, [CDF(1, i/(i+1)) for i in range(10^6)]) > sage: %time c1 = a.conjugate() > Wall time: 4.39 s > sage: %time c2 = a.parent()(a.numpy().conj()) > Wall time: 11.8 ms > Yes, while there are some valid vector-mutating algorithms, I think they would usually arise within linear algebra modules. On user level, a mutable vector is usually just an array and would be served perfectly by a numpy (or python) array. -- 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 view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/ab3a82dc-4150-446c-8934-ec8a32d885ben%40googlegroups.com.
