2013 twenty-four merry days of Perl Feed

Bashing Perl5

hub, bash - 2013-12-21

In the past 6 months, I've worked on a lot of things in a lot of languages. Being an Acmeist, this shouldn't seem suspect. What may come as a surprise, is that my main go-to programming language during this time has been Bash! Bash? Is that even a programming language?

Bashing Ingy

Let me introduce you to my favorite new tool: git-hub. Install this tool and now you can do all your day-to-day GitHub interaction using git commands.

Who is Ingy döt Net anyway?

    $ git hub user ingydotnet

Who cares?

    $ git hub followers ingydotnet

What good is he?

    $ git hub repos ingydotnet

I'm confused…

    $ git hub help

OK, tell me about the git-hub repo?

    $ git hub repo ingydotnet/git-hub

Fork that Bash stuff!

    $ git hub fork ingydotnet/git-hub

Clone me down, Scotty!

    $ git hub clone git-hub  # your fork

Hmmm, I like git-hub!

    $ git hub star ingydotnet/git-hub

The git-hub command is even chainable. Want to follow all the members of the perl6 GitHub organization, and star all their repos?

    $ git hub members perl6 --all --raw | git hub follow -
    $ git hub repos perl6 -ar | git hub star -

With the --raw flag, git-hub prints just the userid or repo-spec, then the '-' flag tells the command to read each line of STDIN and replace the '-' with it. Voilà!

This is just a glimpse of the GitHub things you can do. Give it a spin.

WTB? (What/Why The Bash?)

This amazingly simple/powerful git command is written in Bash. If you are familiar with how Git works, you know that it's basically a huge collection of subcommand plugins with names like git-clone. Try running this command:

    $ ls `git --exec-path`

You should see well over 100 commands. They can be written in any language. In fact, in the early days they were mostly written in Perl5! As Git became more popular, much of the Perl was ported to C. You'll notice that most of those files are symlinks to the git executable. This means that while git-clone is written in C, you could simply replace that symlink with a Haskell program of your own crafting.

In May 2013, I became interested in a replacement for git-submodule, called git-subtree, but soon noticed it had lots of problems too, so I started writing git-subrepo. The subtree command was written in Bash, so I started writing the subrepo command in Bash. I put the subrepo command on hold and started working on the git-hub project. For fun, I decided to see if it was possible to write it in Bash (since I could easy revert to Perl5 if it wasn't).

Was it? It was!

When I try out a new programming language, I generally just think in terms of my favorite programming idioms, rather than language specific idioms. When I get stuck writing something, I google the idiom and read the (almost always StackOverflow) explanation. It starts off slow, but fairly soon I'm getting stuff working, and eventually I feel like a Rock Star.

With Bash, every idiom I needed was there, albeit often in a ridiculous syntactic form, compared to Perl5 (or its hipster cousin Ruby).

The Good, The Bad, and the WTF!

Let's look at some Modern Bash. This is the part of the git-hub code where the subcommands are written. For example, here's the 2 functions for the git hub user command:

command:user() {
  get-args ?user:get-user
  api-get "/users/$user"
  msg_404="User '$user' not found."

ok:user() {
    login type name email blog location company bio
    followers following public_repos public_gists

Quite often when people see this code for the first time they don't believe it's plain old Bash. Well, it is! (And it isn't!). The code above has been heavily refactored, with some Ingy DSL sprinkled on top. Let's review this snippet:

This is the definition of 2 functions. The : is just part of the function name. Bash is quite liberal on function name characters, while quite strict on variable names. Since Bash lacks any concept of hierarchical namespacing, the : lets us fake it nicely. The () parens are required but useless (ie nothing ever goes inside them).

The get-args is just another internal function. I've been using '-' instead of '_' because, you know, it feels more modern. Note that the argument has no quotes. Being a lover of YAML, I avoid quotes when I can. In Bash, commands go through a process called Parameter Expansion, in which whitespace is King. I only use quotes when a string has variable interpolation and the variable in question might have whitespace that would mess up the expansion.

The fields=(…) is how you initialize an array. It's so elegant, and yet this elegance is a Bash rarity. Here's another sexy idiom:

fields+=(foo bar)   # push 2 elements onto an array variable

What if we wanted to print the contents of the array or its size?

echo $fields       # Nope. Only prints 'login'.
echo ${fields[@]} # Correct (and ugly)
echo ${#fields[@]} # Print the array size (so obvious, right?)

As you can see, the Ugly is never far away. What if we want to join the array into a string where the elements are separated by a :?

printf -v string "$(IFS=':'; eval echo "\"${fields[*]}\"")"
echo $string


Which Advent Calendar is this anyway?

You ♥ Perl5. Why is Ingy Bashing you? (tis the season to be Merry!!!)

As I swam through the enigmatic waters of Bash, it became very apparent to me, where Perl got its roots. There was a consistent string of Aha! moments, where I would see where Larry got his inspirations and then cranked them each up to 11.

A friend pointed out to me that he read somewhere, that Perl was originally an attempt to do Shell programming in a single process. This was after I noticed that the simple git hub user command spawned ~200 subprocesses! I guess subprocesses are cheaper now than in the 80s.

Which reminds me, I only realized yesterday that Bash 1.0 was released 2 years after Perl 1.0. Bash is the FSF's free offering of the Bourne shell (from which all the great Unix shells descended). It turns out that Bash got more ideas from Perl than the reverse. Perl 1.0 only had sh, csh and ksh to start from.

I have a book's worth of interesting Perl5↔Bash stuff to tell you. Maybe I should write one.

Bash Into the Future

Why is Bash of interest as a programming language? Well for one, it's fairly ubiquitous. This lets you write things that will cause the least pain for the most general audience. But the same can be said of Perl5. It seems to be installed by default as much as Bash.

Well, there's a couple things that Shell is way better at than Perl/Ruby/Python etc: IPC and system interaction. These things are so naturally ingrained into Shell, that they often go unnoticed, and they make Shell a much more natural language for writing more Shell commands, UNTIL… you need a module like, say JSON or YAML.

The real problem with programming in Bash is there's no CPAN. Fortunately that's a solvable problem! Just a simple matter of programming. I realized early into git-hub development that I needed a complete JSON parser. Not only that, Bash has no data structures to parse JSON into. I solved both problems by writing json-bash. I also needed to write tests. Check out test-more-bash. All these obscure idioms could be packaged into a nice DSL, to make Bash hacking more accessible. Introducing Bash+.

Hmm… This is turning into a nice little collection. Hmm… Where to host all these modules, and start the Bash revolution? I've owned the domain bpan.org for many years now. Hmm… stay tuned!!!

Wrapping it Up

The range of influence that Perl5 has had on modern computing is vast. Bash is a great language to simultaneously see Perl's roots and its contributions in one place. I encourage you to dive into other programming languages, and swim around. If you're like me, it will grow your love for Perl. You'll have parts of Perl culture to grace on other communities, and you'll find things to bring back home. Because if you are really like me, you call Perl your family.

Merry Christmas :-♥

Gravatar Image This article contributed by: Ingy Döt Net <ingy@ingy.net>