Support other element types

[superbobry]

There're quite a few data types we could support:

• byte
• short
• int
• long
• float

with int probably being the most useful (and common) one. However, I'm not sure how to go about code generation. The usual stringly-typed approach looks too error-prone.

[dievsky]

Float also looks important, for those users with stringent memory requirements.

Seeing as Kotlin is a strictly-typed language, we would have to eschew NumPy's approach (store the actual element type as a string property) and instead provide a separate class (hierarchy) for each element type (F32Array for floats, I32Array for ints etc.) We certainly can't use generics, because that would just box everything all the time.

The code generation, however, does present a problem. I'm tempted to invent a C++ template / Kotlin crossover which would automatically generate Kotlin code based on a template. Without this, our code is bound to get really WET or even WETTER (Write Everything Three Times and Enjoy Refactoring).

[dievsky]

https://github.com/JetBrains-Research/kinference-primitives is a C++ template / Kotlin crossover which automatically generates Kotlin code based on a template.