Parsing text in Ruby, part 1

Parslet transforming my world

March 15, 2023 Ā· Felipe Vogel Ā·

I made a new Ruby gem: Litter. Itā€™s for avid trash-picker-uppers like me who keep a log of interesting litter that I collect.

(Surely there are other people who do thatā€¦ *does some googling* See? Iā€™m not alone!)

But I had another reason to make this odd gem: I wanted to explore a more structured approach to text parsing than what Iā€™ve come up with on my own so far. So I used the Parslet gem, and I thought it would be worth writing this post on how it went.

Regular expressions: not the problem

Hereā€™s the backstory. Iā€™m building another gem called Reading that parses my CSV reading log. It has a custom-built parser that is quite messy, for reasons that I couldnā€™t articulate until just now when I tried Parslet.

At first I thought the messiness came from the numerous regular expressions in my homespun parser. The well-known saying comes to mind:

Some people, when confronted with a problem, think ā€œI know, Iā€™ll use regular expressions.ā€ Now they have two problems.

ā€“ Jamie Zawinski

Itā€™s true that some of my regular expressions are long and hard to read, but I donā€™t think regular expressions are the problem because even if I broke them up and made them easier to read, that parser code would still be a mess.

The problem with my Reading parser, I now realize, is this: it mixes up parsing and transformation rather than separating them into two steps.

Hold up, what is ā€œtransformationā€ and how is it different from parsing? I didnā€™t know either, before I tried Parslet. So letā€™s take a look at Parslet to find out!

Discovering Parslet

In the book Text Processing with Ruby I found the structured parsing tool that I was looking for: Parslet. Below is an example similar to the one in the book.

A lot of the syntax is self-explanatory, and for the rest you can refer to Parsletā€™s Get Started, Parser, and Transformation guides.

require "parslet"

# An example config file to be parsed.
INPUT = <<~EOM.freeze
  name = Felipe's website
  url = http://fpsvogel.com/
  cool = true
  post count = 37
EOM

# The output after parsing and transforming.
OUTPUT = {
  name: "Felipe's website",
  url: "http://fpsvogel.com/",
  cool: true,
  post_count: 37,
}

# Parses a string into a tree structure, which we'll then transform
# into the above output. (See the code at the bottom.)
class MyParser < Parslet::Parser
  rule(:whitespace) { match('[\s\t]').repeat(1) }
  rule(:newline) { str("\r\n") | str("\n") }

  # e.g. "name" or "url" in the example above.
  rule(:key) { match('[\w\s]').repeat(1).as(:key) }
  rule(:assignment) { str('=') >> whitespace.maybe }
  # e.g. "Felipe's website" in the example.
  # All characters until the end of the line.
  rule(:value) { (newline.absent? >> any).repeat.as(:string) }

  rule(:item) { key >> assignment >> value.as(:value) >> newline }

  rule(:document) { (item.repeat).as(:document) >> newline.repeat }

  root :document
end

# Transforms a parsed tree into a hash like OUTPUT above.
class MyTransform < Parslet::Transform
  rule(string: simple(:s)) {
    case s
    when "true"
      true
    when "false"
      false
    when /\A[0-9]+\z/
      s.to_i
    else
      s.to_s
    end
  }

  rule(key: simple(:k), value: simple(:v)) {
    [k.to_s.strip.gsub(" ", "_").to_sym,
      v]
  }

  rule(document: subtree(:i)) { i.to_h }
end

parsed = MyParser.new.parse(INPUT)
hash = MyTransform.new.apply(parsed)

puts hash == OUTPUT
# => true

To summarize, a two-step process happens here: first the INPUT string is parsed into an intermediate tree structure, and then the tree is transformed into the simpler hash as in OUTPUT.

Neat! This looks a lot cleaner than if I had mixed parsing and transformation together like I did in my Reading parser.

Impressed by this tidiness, I proceeded to build my litter log parser with Parslet.

The part where I wanted to run crying back to regular expressions, until I learned to implement my parser incrementally

My enthusiasm was curbed as soon as I wrote my first attempt at a parser aaaandā€¦ I got a Parslet::ParseFailed error. It did tell me the input line where the problem occurred, but that doesnā€™t help when there are many rules at play in one line and I donā€™t know which of them needs to be adjusted. I was stumped.

This happened several times until I realized that instead of writing a bunch of rules and then testing them out together, I have to write one rule at a time, or even one bit of a rule at a time, and examine the output at each step. That way, if I get an error then I know itā€™s because of the one change that I just made.

Takeaways

In the end, my parser and transformation are definitely easier to understand than if Iā€™d winged it and built an ad-hoc parser that (as before) doesnā€™t separate transformation into a separate step.

In the tests you can see example input and example output. Considering how different the input and output are, the amount of code that I had to write is fairly small.

You may have noticed that my transformation class doesnā€™t use Parslet rules. Thatā€™s because Parslet::Transform works best when a parsed tree is very predictable in its structure, and when the basic structure of the tree doesnā€™t need to change. To quote Parsletā€™s ā€œTransformationā€ doc:

Transformations are there for one thing: Getting out of the hash/array/slice mess parslet creates (on purpose) into the realm of your own beautifully crafted AST classes. Such AST nodes will generally correspond 1:1 to hashes inside your intermediary tree.

In my case, I needed to radically change the structure of the output (grouping item occurrences by item instead of by date of occurrence), so it made sense to iterate over the parsed output in my own way.

Next steps

After this trial run with Parslet, Iā€™m considering using it in my larger project Reading, where it could replace my ad-hoc parsing code. It would take a lot of work to replace what amounts to the majority of the code in that gem, but it might be worthwhile for a couple of reasons:

And, well, doing more Parslet (or taking a similar parse-and-transform approach) is the thing Iā€™m most interested in right now, and I think that counts for something in my open-source project that no one uses besides me šŸ˜‚

Before you go, hereā€™s a particularly trashy spot in the park where Iā€™ve begun cleaning up. (That is a mattress in the upper-right corner. It will definitely find its way into my litter log.)

trash strewn along a stream in my neighborhood park

šŸ‘‰ Next: Parsing text in Ruby, part 2 šŸ‘ˆ Previous: Switching to Linux šŸš€ Back to top