Almost all haskellers end up, some day, having to write a parser. But then, that’s not really a problem because writing parsers in Haskell isn’t really annoying, like it tends to be elsewhere. Of special interest to us is attoparsec, a very fast parser combinator library. It lets you combine small, simple parsers to express how data should be extracted from that specific format you’re working with.

Getting our feet wet with attoparsec

For example, suppose you want to parse something of the form |<any char>| where <any char> can be… well, any character. We obviously only care about that precise character sitting there – once the input is processed, we don’t really care about these | anymore. This is a no-brainer with attoparsec.

Here we go, we have our parser. If you’re a bit lost with these combinators, feel free to switch back and forth between this article and the documentation of Data.Attoparsec.Text.

This parser will fail if any of the 3 smaller parsers I’m using fail. If there’s more input than just what we’re interested in, the additional content will be left unconsumed.

Let’s now see our parser in action, by loading it in ghci and trying to feed it various inputs.

First, we want to be able to type in Text values directly without using conversions functions from/to Strings. For that reason, we enable the OverloadedStrings extension. We also import Data.Attoparsec.Text because in addition to containing char and anyChar it also contains the functions that let us run a parser on some input (make sure attoparsec is installed).

Data.Attoparsec.Text contains a parse function, which takes a parser and some input, and yields a Result. A Result will just let us know whether the parser failed, with some diagnostic information, or if it was on its way to successfully parsing a value but didn’t get enough input (imagine we just feed "|x" to our parser: it won’t fail, because it looks almost exactly like what we want to parse, except that it doesn’t have that terminal '|', so attoparsec will just tell us it needs more input to complete – or fail), or, finally, if everything went smoothly and it actually hands back to us a successfully parser Char in our case, along with some possibly unconsumed input.

Why do we care about this? Because when we’ll test our parsers with hspec-attoparsec, we’ll be able to test the kind of Result our parsers leaves us with, among other things.

Back to concrete things, let’s run our parser on a valid input.

That means it successfully parsed our inner 'x' between two '|'s. What if we have more input than necessary for the parser?

Interesting! It successfully parsed our 'x' and also tells us "hello world" was left unconsumed, because the parser didn’t need to go that far in the input string to extract the information we want.

But, if the input looks right but lets the parser halfway through completing, what happens?

Here, the input is missing the final | that would make the parser succeed. So we’re told that the parser has partially succeeded, meaning that with that input, it’s been running successfully but hasn’t yet parsed everything it’s supposed to. What that Partial holds isn’t an just underscore but a function to resume the parsing with some more input (a continuation). The Show instance for parsers just writes a _ in place of functions.

Ok, and now, how about we feed some “wrong data” to our parser?

Alright! Equipped with this minimal knowledge of attoparsec, we’ll now see how we can test our parser.

Introducing hspec-attoparsec

Well, I happen to be working on an HTML parsing library based on attoparsec, and I’ve been using hspec for all my testing needs these past few months – working with the author surely helped, hello Simon! – so I wanted to check whether I could come up with a minimalist API for testing attoparsec parsers.

If you don’t know how to use hspec, I warmly recommend visititing, it is well documented.

So let’s first get the boilerplate out of our way.

And sure enough, we can already get this running in ghci (ignore the warnings, they are just saying that we’re not yet using our parser or hspec-attoparsec), although it’s quite useless:

Alright, let’s first introduce a couple of tests where our parser should succeed.

We’re using two things from hspec-attoparsec:

  • (~>), which connects some input to a parser and extracts either an error string or an actual value, depending on how the parsing went.
  • shouldParse, which takes the result of (~>) and compares it to what you expect the value to be. If the parsing fails, the test won’t pass, obviously, and hspec-attoparsec will report that the parsing failed. If the parsing succeeds, the parsed value is compared to the expected one and a proper error message is reported with both values printed out.

Running them gives:

If we modify our first test case by expecting 'b' instead of 'a', while still having "|a|" as input, we get:

Nice! But what else can we test? Well, we can test that what we parse satisfies some predicate, for example. Let’s add the following to spec:


And we get:

Great, what else can we do? Well, sometimes we don’t really care about the concrete values produced, we just want to test that the parser succeeds or fails on some precise inputs we have, because that’s how it’s supposed to behave and we want to have a way that changes in the future won’t affect the parser’s behavior on these inputs. This is what shouldFailOn and shouldSucceedOn are for. Let’s add a couple more tests:


And we run our new tests:

I think by now you probably understand how to use the library, so I’ll just show the last useful function: leavesUnconsumed. This one will just let you inspect the unconsumed part of the input if there’s any. Using it, you can easily describe how eager in consuming the input your parsers should be.

Right now, hspec-attoparsec will only consider leftovers when the parser succeeds. I’m not really sure whether we should return Fail’s unconsumed input or not.


The code lives at, the package is on hackage here where you can also view the documentation. A good source of examples is the package’s own test suite, that you can view in the repo. The example used in this article also lives in the repo, see example/. Let me know through github or by email about any question, feedback, PR, etc.