Broadcasting? #31
Description
We haven't implemented array broadcasting yet, and I don't think we have actually talked much about it. I'm not opposed to the idea, but in my years of using numpy, I don't believe I've ever fruitfully used it. Additionally, it is quite complicated to reason about as a user.
Is broadcasting actually important?
Pros:
- Might increase performance. The current ways I'd reccommend for emulating broadcasting are:
- using a tiling iterator (performant, but not vectorizable and relatively involved, probably takes >5 lines)
- using #build to manually construct the result (not performant, but pretty easy, and clear to understand)
- Terse
Cons:
- Increases code complexity
- Provides magic that a user might not expect
- Can hide bugs by failing to issue a ShapeError
- Terse