Hacker News Reader

Comment by jp0d | Hacker News Reader

Controversial opinion here! I don't think people who use languages like Julia/R/Python etc worry about most of the computer science side of things. Most of them are data scientists/analysts, statisticians and from pure sciences. Primarily because the availability of wide range of libraries that makes their lives a lot easier. Also many aren't too concerned about performance of the language either. It's indeed fascinating that the author's first language was Julia. It'd be interesting to know why the author chose Julia as the first language.

View on HN

I don’t find that to be true with Julia. Keep in mind that Julia is intended to not only replace python and R but also C++ in a science setting. The academic programmers who write the high performance code parts usually do this in C/C++ and Fortran and thus tend to be quite familiar with a lot of low level computer science stuff.

Looking at generated assembly code for a Julia function is a pretty normal thing to do. I am an old C/C++ developer and find myself very much at home with Julia.

Sure I have used Python and Ruby before but those always make decisions around memory layout and code generation inaccessible.

Thinking about Julia as yet another Python, R, Ruby or a JavaScript is just wrong. It doesn’t really fit existing categories well.

I mean, the best way to think about it is as a replacement for MATLAB, except with better support for general purpose computing. MATLAB the language syntax is optimized for double precision matrix math - but MATLAB is interpreted with a JIT and extremely productive IDE, so actually really easy to use, except for the part where it costs $2000 per toolbox per year. It also interops fairly seamlessly with Java since it runs on the JVM.

Mathworks: the company that I wouldn't pay paying money to, if it wasn't such a ridiculously large sum. I don't have any issue paying for Jetbrains IDEs, but it's an order of magnitude smaller than Mathwork's yearly take.

Because I heard the good word of our Lord and Savior David Sanders [1] in my computational physics class and I've renounced the devious ways of Fortran since.

[1]: https://www.youtube.com/watch?v=gQ1y5NUD_RI

Concretely, I joined the Julia slack and ran into a very welcoming community of people who helped me learn. Julia's like that so far.

Yes, I agree many data science people don't dive into internals - but for some reason these low-level details keep fascinating me so I don't mind some of the pain.

I don't know, there is a lot of us who moved to Julia without doing R, or using Python as a stats language.

There is a lot of people wanting to use things like JuMP and handle numeric types where a single value is things like probability curves, and you want to use regular libraries with them, without having to rewrite those libraries.

Julia is really it's own beast, way way beyond R / Python.

It is friendly towards R dev's but, people who normally had to hack stuff in Fortran, C++, etc love it.

Performance does matter in HPC but they tune for performance in a different way, not by the developer being smart really but by the language (read fortran) being dumb, in particular, easy to slice up for the compilers that are honestly decades old.

> not by the developer being smart really but by the language (read fortran) being dumb,

That's definitely a valid approach, but it's far from the only one employed in HPC.

Modern Fortran is quite smart, including modules, OOP and generics support.

Also another wannabe replacement, Chapel is relatively smart, trying to mix Fortran productivity for HPC code with Rust like memory management.

I really admire "Julia" for not being that, for basically saying "yes, this is a programming language by and for people focused on other things because just because you aren't in the bike industry means you are best off commuting on a 1960s Huffy". At the same time, I wish a few more people had bothered to "be the bridge", because I think it's worse than I could be for not being super up to date the PL literature.

So love the idea, but regret the execution.

Reply on news.ycombinator.com