Parsing text with clojure.spec – The library that keeps on givin'

This week I’ve been working on JUXT’s tick library.

Tick is a time library, and has a few playful functions for working out dates like Easter Sunday. But as part of a major overhaul of the library, towards a serious production-ready library, we’ve decided it needs to read holiday dates from official information sources (not all holidays can be computed!) and for that, it needs to be able to read the iCalendar format.

iCalendar is specified in RFC 5545, and like many RFCs, grammar rules are defined in a syntax known as Augmented Backus-Naur Form (ABNF), which is a lot more intuitive than it sounds.

iCalendar files are made up of a set of 'content' lines defined by the following rule:

contentline = name *(";" param ) ":" value CRLF

This tells us that each line has to begin with a name, optionally followed by some parameters, then a colon followed by a value, ending with a line-ending.

Of course, we now need to know precisely what names, parameters and values are. These are also defined for us (but for the purposes of this article I’ve simplified somewhat):

name = iana-token
iana-token = 1*(ALPHA / DIGIT / "-")
param = param-name "=" param-value
param-name = iana-token
param-value = iana-token
value = iana-token

Parsing

Having split up our iCalendar file into content lines (and there are some folding rules in doing that which we’ll ignore), we now need to parse them.

I’d like our solution to have the following properties:

  1. Be easy to understand and update (my simplifications aren’t going to go the distance)

  2. Be reasonably performant

  3. Be done by the time I get to the JUXT office (I’m writing this on my morning bus journey)

We could start by trying Regular Expressions, which are well provided for in Clojure and Java. But regexes don’t cope well with any degree of complexity. Our 'zero-or-more parameters' rule is going to trip us up.

At this point, I’d usually roll up my sleeves and be reaching for tools like instaparse in Clojure (or Antlr in Java) and paging-in all that Computer Science 101 stuff about LALR(1) context-free grammars, blah blah blah. But I’m on the bus and have rather intermittent Internet connectivity. Is there a quicker/better way?

Turns out, tick already has a dependency on clojure.spec, so can I use that?

Let’s start creating a spec for iana-token.

(require '[clojure.spec :as s])

(s/def ::iana-token
  (s/+ (set "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ01234567890-"))

;; Let's try it
(s/conform ::iana-token (seq "DTSTART")) => [\D \T \S \T \A \R \T]
(s/conform ::iana-token (seq "&64*")) => :clojure.spec.alpha/invalid

Gosh, that wasn’t so hard.

Now let’s see if we can hand-write our rule for contentline:

(s/def ::contentline
  (s/cat
    :name ::iana-token
    :params (s/*
              (s/cat
                :semi #{\;}
                :param-name ::iana-token
                :equals #{\=}
                :param-value ::iana-token))
    :colon #{\:}
    :value ::iana-token))

Notice how readable and intuitive that code is.

Let’s try that rule with some a real iCalendar contentline:

(s/conform
  ::contentline
  (seq "DTSTART;TZID=US-EAST:20180116T140000"))
=>
{:name [\D \T \S \T \A \R \T],
 :params
 [{:semi \;,
   :param-name [\T \Z \I \D],
   :equals \=,
   :param-value [\U \S \- \E \A \S \T]}],
 :colon \:,
 :value [\2 \0 \1 \8 \0 \1 \1 \6 \T \1 \4 \0 \0 \0 \0]}

clojure.spec has broken the string up into just the pieces we need. For our parameters, it’s given us a collection, because there could be a few of them.

As a quick example, let’s extract the value of first parameter.

(apply str (get-in … [:params 0 :param-value]))
=> "US-EAST"

Voila!

Conclusion

The fact that clojure.spec—which was designed for a different purpose—seems so adept in the area of parsing is a surprise, but it’s a sign that there’s a lot of power in this little library. More evidence that where many languages veer off into the minutia, Clojure gets the job done.

The performance on my laptop isn’t bad (~500µs), and will parse a 5000-line iCalendar file in couple of seconds, which is within my requirements.

Looks like I’m done, and my bus has arrived!

(With thanks to @thegeez for this gist which inspired me to try clojure.spec to solve this problem.)

Published: 2018-01-17

Privacy policy

Extraterrestrial intelligence oly
by Oliver Hine
Published: 2016-11-08
A case study of building the infrastructure for an electric powered bikes scheme
Published: 2015-10-29
It's very possible dmc
by Dominic Monroe
Published: 2016-12-13