Administrative Scripting with Julia

Note that this tutorial is still in process. The intended order, after the introduction is:

• files
• CLI
• filesystem
• processes
• regex (still being written)

However, each part should theoretically stand on its own to some extent.

Those parts which have been written should nonetheless be considered drafts for the moment, but you may still find them useful.

Introduction

If you know anything about Julia, you probably know it's an interactive language which is gaining popularity for numeric computing that competes with the likes of R, Matlab, NumPy and others. You've probably also heard that it compiles to machine code at runtime¹ with an LLVM JIT and can often be optimized to within a factor of two of C or Fortran.

Given these promises, it's not surprising that it's attracted some very high-profile users.

I don't do any of that kind of programming. I like Julia for other reasons altogether. For me, the appeal is that it feels good to write. It's like all the things I like from Python, Perl and Scheme all rolled into one. The abstractions, both in the language and the standard library, just feel like they are always hitting the right points.

Semantically and syntactically, it feels similar to Python and Ruby, though it promotes some design patterns more common in functional languages and doesn't support classical OO patterns in the same way. Instead, it relies on structs, abstract types, and multiple dispatch for its type system. Julia's metaprogramming story is simple yet deep. It allows operator overloading and other kinds of magic methods for applying built-in syntax features to your own types. If that isn't enough, it has Lisp-like AST macros.

Finally, reading the standard library (which is implemented mostly in very readable Julia), you see just how pragmatic it is. It is happy to call into libc for the things libc is good at. It's equally happy to shell out if that's the most practical thing. Check out the code for the download function for an instructive example. Julia is very happy to rely on PCRE for regular expressions. On the other hand, Julia is fast enough that many of the bundled data structures and primitives are implemented directly in Julia.

While keeping the syntax fairly clean and straightforward, the Julia ethos is ultimately about getting things done and empowering the programmer. If that means performance, you can optimize to your heart's content. If it means downloading files with curl, it will do that, too!

This ethos fits very well with system automation. The classic languages in this domain are Perl and Bash. Perl has the reputation of being "write only," and Bash is much worse than that! However, both are languages that emphasize pragmatism over purity, and that seems to be a win for short scripts. Julia is more readable than either of these, but it is not less pragmatic.²

This tutorial follows roughly the approach of my Python tutorial on administrative scripting and may refer to it at various points. Note that I've been using Linux exclusively for more than a decade and I'm not very knowledgable about Windows or OS X. However, if people wish to contribute content necessary to make this tutorial more compatible with those platforms, I would be very happy to learn.

Why You Shouldn't Use Julia for Administrative Scripts

It's just a bad idea!

• Julia has a fat runtime and it has a human-perceptible load time on a slower system. For the kinds of intensive problems it's targeted at, this is nothing. On a constrained server or an embedded system, it's bad.

• Julia's highly optimizing JIT compiler also takes a little time to warm up. There are ways to precompile some things, but who wants to bother for little scripts? The speed of the compiler is impressive for how good it actually is, but it's not instant.

  Note: in recent versions of Julia, there is a --compile=min option which can be used to force Julia to do less code specialization. Combining this with -O0 will reduce JIT warmup time, though method resolution will still take time in some instances.

The above are reasonable arguments against using Julia on a certain class of servers. However, none of this stuff really matters on a PC/workstation running an OS with current packages. If your system can run a modern web browser, Julia's runtime is a pittance.

If you already want to learn Julia, which there are many good reasons to do, writing small automation scripts is a gentle way to become acquainted with the basic features of the language.

The other reason you might want to try administrative scripting in Julia is because the abstractions it provides are surprisingly well suited to the task. Translating a Bash script to Julia is very easy but will automatically make your script safer and easier to debug.

One final reason to use Julia for administrative is that it means you're not using Bash! I've made a case against Bash for anything but running and connecting other processes in Bash in my Python tutorial. In short, Bash is great for interactive use, but it's difficult to do things in a safe and correct way in scripts, and dealing with data is an exercise in suffering. Handle data and complexity in programs in other languages.

Learning Julia

This tutorial isn't going to show you how to do control flow in Julia itself, and it certainly isn't going to cover all the ways of dealing with the rich data structures that Julia provides. To be honest, I'm still in the process of learning Julia myself, and I'm relying heavily on the official docs for that, especially the "Manual" section. As an experienced Python programmer, the interfaces provided by Julia feel very familiar, and I suspect the feeling will be similar for Ruby programmers. For us, becoming productive in Julia should only take a few hours, though there are rather major differences as one progresses in the language.

For a quick introduction to the language, the learning page has some good links. The Intro to Julia with Jane Herriman goes over everything you'll need to know to understand this tutorial. If you choose to follow this tutorial, you will be guided to log into juliabox.com, but you don't need to unless you want to. You can download and run the Jupyter Notebooks locally if you wish, and you can also simply follow along in the Julia REPL in a terminal.

The Fast Track to Julia is a handy cheatsheet if you're learning the language

Anway, let's get straight on to files.

---

1. recent versions of Julia do more caching of generated machine code.↩

2. This is not to fault the creators of Perl or Bourne Shell. They are much older langauges, and all dynamic languages to come after, including Julia, have followed in their footsteps. Later languages learned from their problems, but they also learned from what they did right, which was a lot!↩