how to "turn off" the term presence lunr setting for queries that start with a hyphen -

[flynneva]

After doing some digging on why search queries that start with a hyphen (see the Term Presence section) dont seem to return the expected results I was wondering how to "turn off" this feature of lunr so that when a user queries --some-argument they get the expected result?

I noticed that on the mkdocs site querying for --force returns the expected results so I am sure it is possible!

[flynneva]

so I was able to semi-figure it out through a hacky way using the custom overrides feature of mkdocs to remove the leading - if present before submitting the query to lunr. This has a drawback of now removing the "exclude" feature that leading - are supposed to have...but for our use case I think this solution might work.

here is what i added to our overrides/main.html:

{% block config %}
  {{ super() }}
  <script>
    var __search = {
      transform: function(query) {
        return query
          .split(/"([^"]+)"/g) 
            .map((terms, index) => index & 1
              ? terms.replace(/^\b|^(?![^\x00-\x7F]|$)|\s+/g, " +")
              : terms
            )
            .join("")
            .replace(/"|(^\-+)|(?:^|\s+)[*+:^~]+(?=\s+|$)/g, "")
          .trim()
      }
    }
  </script>
{% endblock %}

[squidfunk]

I noticed that on the mkdocs site querying for --force returns the expected results so I am sure it is possible!

This is because - is used as a separator, so --force is tokenized into force, which is why it's discoverable. I don't think it's possible to craft a regular expression that detects - as a separator in foo-bar, but leaves it as-is in --foo, but I may be mistaken.

[flynneva]

@squidfunk I had already adjusted our separator tokenizer regex to only include a - if it is not followed by a non-whitespace character \-(?!\S) (I actually do it the opposite, so check if it is whitespace and include it \-(?=\s)) very similar to how you recommended to find version numbers in your docs.

I guess an alternative solution would be to switch the above html code to only replace the leading hypens if there are exactly two - that way the original functionality will still be there for excluding terms from results. This approach would work for many command line args but not all I guess.