| Authors: | Stefan Karpinski |
|---|---|
| Time: | 2:00 pm - 2:40 pm |
| Session: | https://thestrangeloop.com/sessions/julia-a-fast-dynamic-language-for-technical-computing |
| Link: | http://julialang.org/ |
Julia is a fast, dynamic language for technical computing. “Technical Computing” is sort of a made up term, but it includes languages like Matlab, Maple, Mathematica, R, SciPy, etc. There are at least forty technical computing languages. Julia had three feature goals: dynamic language, sophisticated parametric type system, and multiple dispatch. Existing languages have some of these, but the combination is somewhat unique in Julia.
Julia can be Matlab-like: code is defined in functions, [shows example]. But you can also write lower-level, non-vectorized code [shows example of their quicksort micro-benchmark]. Julia supports distributed computation, and macros for constructs _like_ distributing computation.
Shows the difference between how Python and C store arrays. C requires that arrays have a single type, so you can store (say) the list of floating point things in contiguous memory. Python lets you do anything in the list, so it stores a list of pointers to the elements. Julia wanted to be able to store the information contiguously in memory, and still have flexibility. This means you can’t change a type once it’s declared: you can’t make it bigger, add fields, etc. [Did I mishear that this means there’s no/limited sub-classing?]
Discussion of how Julia handles multiple dispatch. Uses explicit promotion instead of overloading.
[ Live coding demo of hypothetical Modular Int type. ]
Julia performs very well compared to Python, Matlab, Octave, R. Julia runs on LLVM, so fast to start with, and they’re working on performance.
Julia performs no static type checking.