Shell scripting with Elixir | Hacker News Reader

Shell scripting with Elixir

arathunku.com

140 pointsbo0tzz43 comments

Shell scripting with ideas from Elixir, chiefly, functional programming style would be profitable too. As an avowed (mostly) functional programmer, I do it FP style [1].

Of course, one can take a Shell script only so far. Eventually a Real Programming Language is the inescapable (pun intended) conclusion. [Insert "Always was" meme].

[1] I wrote about it here.

https://www.evalapply.org/posts/shell-aint-a-bad-place-to-fp...

That was so cool. Kind of an eye-opener. Bookmarked, thank you!

Thank you for the kind remark. It so happened that I picked up shell scripting on the job at a workplace full of FP nerds. Pipelined code felt eerily familiar. Then at some point I read that Archmage McIlroy, a mathematician by training, invented the pipe. That was a lightbulb moment; a big reason why Functional style works as well as it does. Pipes are a mathematical idea.

That whole series of posts is basically a glorified bookmark for yours truly. Tactics and tricks accumulated over the years. I refer back to bits and bobs from time to time. Writing them was a lot of, ah, fn.

This Bash library [1] is very handy to make your scripts more functional.

[1] https://github.com/ssledz/bash-fun

I feel like I've seen this one before. Thanks for bubbling it back up in my brain.

I think `map`, `filter`, and `reduce` (or `accumulate`) are automatic in pipelines; and have usually been enough for me. Still, it has a bunch of stuff one can pinch in a hurry. Starred!

Shell scripting in Elixir is like leaf-blowing your driveway with a helicopter.

Awesome, but too expensive to be within reach of the ordinary person and more harmful to the environment compared to alternative methods?

elevator leaf blowing might be harmful to the environment but whats the argument about elixir in this analogy?

I have no idea what parent is trying to say with their analogy, I'm trying to tease out some meaning from what they said. If anything, Elixir would be more energy efficient than bash as far as I can tell.

> Elixir would be more energy efficient than bash as far as I can tell

For long running scripts perhaps. BEAM idles at ~50Mb, and startup is often more than 500ms.

Eg:

  $ time ./hello
  hello world

  real    0m0,009s
  user    0m0,003s
  sys     0m0,007s
  
  $ time elixir hello.exs
  hello world
  
  real    0m0,755s
  user    0m0,992s
  sys     0m0,304s

Where the first example is a BASH script.

Yes because when you load elixir you're loading the entire gOddamn Erlang VM. Which you might want for things like reliable parallel downloading, but you don't for hello world.

I don't know that you'll get it down to 10 ms, but you can get beam startup a lot faster if you adjust parameters.

The default startup does a lot of things you simply don't need in most shell scripts. Starting a bunch of threads, setting up dist, I can't remember what else off the top of my head.

Even without the shell and with epmd turned off we're still in the ball bark of 200-300ms:

  $ time erl -eval 'io:format("~s~n", [os:cmd("echo hello world")]), halt(0, [{flush, false}]).' -noshell -start_epmd false

hello world

  real    0m0,332s
  user    0m0,461s
  sys     0m0,092s

I spent some time last year trying to get those numbers down but this is as far as I made it. Do you know of any more flags that might help?

Try -boot no_dot_erlang

A couple years ago that got me to 100 ms

https://news.ycombinator.com/item?id=25243246

Today's measurements with FreeBSD 13.2 on the same Pentium G3470

time /bin/sh -c "echo hi"

   hi

   real    0m0.004s
   user    0m0.001s
   sys     0m0.002s

time /usr/local/bin/bash -c "echo hi" hi

   hi

   real    0m0.006s
   user    0m0.007s
   sys     0m0.000s

time /usr/local/lib/erlang26/bin/erl -eval 'io:format("hi~n"), halt(0, [{flush, false}])'

   Erlang/OTP 26 [erts-14.2.1] [source] [64-bit] [smp:2:2] [ds:2:2:10] [async-threads:1] [jit:ns] [dtrace] [sharing-preserving]

   hi

   real    0m0.197s
   user    0m0.178s
   sys     0m0.018s

time /usr/local/lib/erlang26/bin/erl -eval 'io:format("hi~n"), halt(0, [{flush, false}])' -noshell -start_epmd false -boot no_dot_erlang

   hi

   real    0m0.183s
   user    0m0.181s
   sys     0m0.011s

This doesn't look very good, actually! FreeBSD has OTP 21-26 available through pkg, I ran the same thing on each of them several times to warm up, and get (real time):

   21:  81ms
   22:  81ms
   23:  74ms
   24: 129ms
   25: 135ms
   26: 177ms

So maybe there's some new stuff that needs to be avoided now.

It was your original post that set me on course for experimenting with startup times, cheers!

Unfortunately, that final flag doesn't change much on 26 for me. IIRC I dropped a bunch of flags after concluding that no shell and no epmd were the only flags that I noticed improvements with.

Yeah, it looks like it got a lot worse. I don't think it's JIT, because I ran a test with a 32-bit version of OTP 26 (JIT is only available for 64-bit x86 and arm). But I'm not sure what it is.

Send me an email (in profile), and I'll try to find some time to look closer.

If your script needs to do things like reliable HTTP querying and JSON parsing, I would say using Elixir (once you get used to it) won’t be harmful in any way, quite the contrary.

Interestingly, what drove me to Ruby initially is exactly writing scripting tasks in a C#/C++ culture where Ruby would have been considered harmful.

Now Elixir can be used for the same type of tasks, but only better (eg: trivial parallelized downloads!)

F# and C# are very different to C++ in this regard.

Parallel downloads in C# boil down to

    using var http = new HttpClient();

    var urls = new[]
    {
        "http://example.com/file1.jpg",
        "http://example.com/file2.jpg",
        /* ... */
    };
    var files = await Task.WhenAll(
        urls.Select(http.GetByteArrayAsync));

(you can replace '.GetByteArrayAsync' with .GetFromJsonAsync, .GetStreamAsync and more depending on your needs and e.g. forward the incoming data directly to files on the disk or another source)

You should, if you can?

Absolutely. What is life without any fun?

Ahaha, in a good way. Recently I have used Elixir scripts to mass-parallel-download numerous HTTP resources and do some stats on them, and it is exactly the feeling I had.

Suprisingly relevant https://www.youtube.com/watch?v=Ss6VBYLjMJQ

I envision the creation of a GenServer for each and every file on your system, using this to duplicate them by the thousands, then deleting them in a timed contraction over and over and over again. I call this, "the blowcopter".

Still better than the inverse, that lots of scripters are doing; Building a helicopter out of a leaf-blower.

Ever since Elixir 1.12, which introduced the `Mix.install[..]` at the beginning of the script, Elixir has been awesome for this sort of thing. Prior to that, it just wasn't worth it.

I'm surprised the testing worked like that, without trying myself. When I last fiddled with it, I stole a page from how they recommend you do it in Livebook, and had `ExUnit.start(autorun: false)` at the top and `ExUnit.run()` at the bottom. I didn't realize you could just `ExUnit.start()` by itself up front.

For anyone else who's tried an approach like this, have you managed to get doctests to work? I really like them, and think they're especially appropriate where you probably want lots of examples like here, but haven't been able to figure out how to stick them in ad-hoc scripts like this. I get some issue about the code not being compiled and available.

One detail I really like about this is that you can install packages during the script run.

    Mix.install([
     {:req, "~> 0.2"},
     {:floki, "~> 0.30.0"}
   ])

Ruby can do this too:

  require 'bundler/inline'
  
  gemfile do
    source 'https://rubygems.org'
    gem 'json', require: false
    gem 'nap', require: 'rest'
    gem 'cocoapods', '~> 0.34.1'
  end

Can't Python do the same? Like !pip install blahblah

That's not Python syntax. It's a Jupyter extension.

Right, but Python can run command line code in subprocesses too, so it should be possible to run a pip install jupyter command inside Python.

The problem is that installing dependencies for the user or globally will lead to version conflict. You need a separate virtual environment or a custom installation directory for each script to avoid conflicts. There are script runners and copy-pastable code snippets for Python that handle this. They automatically create the virtual environment and install the dependencies when the script runs. I have bookmarked several in https://github.com/stars/dbohdan/lists/python-scripts-with-d... (disclosure: not only is this my list, it also includes my project).

Thanks, I checked out the link, it's quite handy!

You can also use anonymous functions if you find the module syntax a little terse or clunky for shell scripting.

    double = fn a -> add.(a, a) end
    double.(4)

It starts to look a bit like a weird untyped OCaml / F# if you use pattern matching:

    f = fn
      x, y when x > 0 -> x + y
      x, y -> x * y
    end

The more I use bash the more I appreciate it. It has a lot of sharp edges and quirks, but it's coincise, it works basically on all Unix systems and everyone knows (at least a bit of) it.

A lot of attempts to fill this space have large gaps in this aspects. Not sure how this scores on those dimensions.

For up to ~50 lines of script, bash is the most effective thing I found so far. Tried zsh, fish, python, ruby, but for such glue code bash rules.

> The more I use bash the more I appreciate it.

That is true for sure if you write bash scripts quasi daily. However if you don't it becomes very easy to forget, at least for me. That's why I'm the type of person that tries to look out for alternatives in different languages.

I've had and read from cover to cover, Bash, sed, and AWK books, yet when I have to get back to them I still only remeber the most fundamental features that I use every couple of days or weeks.

My rule of thumb for moving from bash to Perl/Python etc is when I need to start implementing multiple arrays and associative arrays to get the job done.

Agree

I found that browsing through the bash source makes remembering things about how to use bash easier. It is simple enough to see the gears at work, comapared to say python interpreter sources.

don't those quirks and sharp edges leave security holes in the script tho?

As I am in the process of learning F# I am considering using it for scripting.

I have read good things about that!

Me too!

For personal stuff I use F# or C# (less and less, these days) for everything unless it's chaining together other command line tools in a way they're explicitly designed for. I've tried PowerShell and can't stand it.

I have 20+ years of .NET, VS IDE and debugger experience I can leverage - it doesn't make sense for me to use anything else.

Reply on news.ycombinator.com