Some regex help from the F# compiler

tl;dr: Make invalid regular expression strings and attempts to access non-existent capture groups a compile-time error, thanks to the Regex type provider.

Standard .NET regex

Say we want to parse out some information from basic Liquid tags, like this:

[Test]
public void GetInformationFromAllSampleTags() {
    const string input =
        @"This is a test. {% sample %}ABC{% endsample %}. Some {% other %} 123 {% endother %} tag.
          {% sample %} DEF {% endsample %}";
    GetSamples(input).ShouldBe(new [] { "ABC", "DEF" });
}

We can give ourselves two problems and implement this using System.Text.RegularExpressions (it looks almost identical in C#, see this gist or footnote1):

// F#
let getSamples s : string seq =
    let re = @"\{%\s*(?<tag>\w+)\s*\%\}(?<contents>(?s:.*?))\{%\s*end\1\s*%\}"
    Regex.Matches(s, re)
        |> Seq.cast<Match>
        |> Seq.filter (fun m -> m.Groups.["tag"].Value = "sample")
        |> Seq.map (fun m -> m.Groups.["contents"].Value.Trim())

F# type provider version

First up we need to add the RegexProvider to our project via nuget: PM> Install-Package RegexProvider.

Now we can rewrite our previous implementation like this:

open FSharp.RegexProvider
type LiquidTagRegex = Regex< @"\{%\s*(?<tag>\w+)\s*\%\}(?<contents>(?s:.*?))\{%\s*end\1\s*%\}" >

let getSamples s : string seq =
    LiquidTagRegex().Matches(s)
        |> Seq.filter (fun m -> m.tag.Value = "sample")
        |> Seq.map (fun m -> m.contents.Value.Trim())

This will compile equivalently to our previous implementation2, but we’ve gained some nice static checks.

We can access the tag and contents capture groups of our match as properties. This isn’t a method_missing-style dynamic lookup – if we rename the group in the regex to (?<notTag>\w+) then we get a compile-time error:

error FS0039: The field, constructor or member 'tag' is not defined

Also neat, if we completely muck up our regex, the compiler will let us know:

error FS3033: The type provider ... reported an error: parsing "[asd" -
Unterminated [] set.

Tests would catch both these errors, but feedback doesn’t get much faster than “as we’re typing the code”, plus we get precise line numbers for the errors as well. It also reduces code noise, dealing directly with the capture group names rather than having to specify particular collection lookups.


  1. An equivalent implementation in C#:
  2. The type provider creates a type with the tag and contents properties, but this type gets erased in the final compiled output, replaced with the Groups accessor code from our original implementation.

Comments