This Week in Mojo 2023-08-11
- New blog post: An easy introduction to Mojo for Python programmers comparing how to calculate Euclidean distance in raw Python, Numpy and Mojo.
We are definitely interested in introducing an
Optional[T] type in the future, but need more traits support built out. Once we have that, we can shift to non-nullable "pointers" by default.
Linear algebra in standard library
We need to figure it out, but I'd prefer to keep the stdlib pretty conventional and focused on "normal" library types like integers, strings, dictionaries, pushing things like tensors and matmul out to a linear algebra package (e.g. see Numpy not being built in). That said, I do think something like that could/should be included in the normal distro.
The Modular Engine will be different, and I expect/hope it to always be the best state of the art implementation of this stuff, but it can do so without trying to hold back other efforts!
Tensor type is new but is also very much a work in progress. We added it because a lot of example notebooks are using very low level programming (effectively malloc+free) and need a simple owned buffer type. The initial idea is that "array like" names would work with arbitrary element types (e.g. you can put a 🐴 or 🐶 object in an array) but that we wanted a more "numerics sounding" collection that would be parameterized on DType.
Mojo is still missing a bunch of type systems features (e.g. traits etc) that will massively impact the shape and structure of the standard library, and as those come in, we'll want to significantly rethink previous decisions. Despite that, we don't want to completely hold back progress on things until all those features are available - we decided it is better to prototype some things, get experience, and be willing to reinvent / replace things over time.
Why not build Mojo on top of Swift?
I'm also a fan of Swift. The major issue with it is that it isn't a member of the Python family. I'm confused why you think that Mojo isn't inspired by Swift, because Mojo certainly is. That said, you're right that building a new thing takes longer than leveraging an existing thing. The reason we're taking this approach is that we're optimizing for quality of result (we want to build the world's best thing) not time to market. I would consider it success if you come to love Mojo more than Swift some day 😄