Not able to search for just numbers in lunr.de

[cadamini]

Probem

In my German and English test documents I have content with the term Port 1234, but searching for 1234 does not work.

Has someone seen the same or a similar problem? Any ideas?

More tests

• Searching for Port 1234 works fine.
• Searching for 1234 in an English document works fine, using the base lunr.js version 2.3.8
• Using this.use(lunr.de) makes it possible to find German umlauts but no numbers anymore.

Test Code

// The required JS files are correctly inserted in the sites head

var idx = lunr(function () {
  this.use(lunr.de)
  this.ref('id')
  this.field('text')

  this.add({
    id: 1,
    text: "Port 1234 is a good port for testing a problem"
  })
})

console.log(idx.search('1234'));
console.log(idx.search('Port 1234'));

Result

[andrewzola]

My Russian docs have the same problem (I use mkdosc, if it have mean)

[cadamini]

Related to the trimmer. If I remove the trimmer completely, it works.

The defined word character defined in line 74 were really strange:

lunr.de.wordCharacters = "A-Za-z\xAA\xBA\xC0-\xD6\xD8-\xF6\xF8-\u02B8\u02E0-\u02E4\u1D00-\u1D25\u1D2C-\u1D5C\u1D62-\u1D65\u1D6B-\u1D77\u1D79-\u1DBE\u1E00-\u1EFF\u2071\u207F\u2090-\u209C\u212A\u212B\u2132\u214E\u2160-\u2188\u2C60-\u2C7F\uA722-\uA787\uA78B-\uA7AD\uA7B0-\uA7B7\uA7F7-\uA7FF\uAB30-\uAB5A\uAB5C-\uAB64\uFB00-\uFB06\uFF21-\uFF3A\uFF41-\uFF5A";

Translates to:

ʸˠ ˤᴀ ᴥᴬ ᵜᵢ ᵥᵫ ᵷᵹ ᶾḀ ỿⁱⁿₐ ₜKÅℲⅎⅠ ↈⱠ ⱿꜢ ꞇꞋ ꞭꞰ ꞷꟷ ꟿꬰ ꭚꭜ ꭤff stＡ Ｚａ ｚ

Potential solution:

lunr.de.wordCharacters = "A-Za-züÜÄäÖöß0-9";

[khawkins98]

lunr.de.wordCharacters = "A-Za-züÜÄäÖöß0-9";

I noticed the German support was also breaking * wildcard support, this also fixes that.

[pizaranha]

I was facing the same issue No results for numeric searches. Then I found that adding '\0-9' at the end of line 74 that will include numeric searching.

lunr.es.wordCharacters = "A-Za-z\xAA\xBA\xC0-\xD6\xD8-\xF6\xF8-\u02B8\u02E0-\u02E4\u1D00-\u1D25\u1D2C-\u1D5C\u1D62-\u1D65\u1D6B-\u1D77\u1D79-\u1DBE\u1E00-\u1EFF\u2071\u207F\u2090-\u209C\u212A\u212B\u2132\u214E\u2160-\u2188\u2C60-\u2C7F\uA722-\uA787\uA78B-\uA7AD\uA7B0-\uA7B7\uA7F7-\uA7FF\uAB30-\uAB5A\uAB5C-\uAB64\uFB00-\uFB06\uFF21-\uFF3A\uFF41-\uFF5A\0-9";

I think it could be a config option in the future.

[blackwidow207]

As of ES6 regexp in JavaScript now supports the unicode flag, so pretty sure this can be used to simplify the trimmer function for all languages when creating the search index. Some of the language implementations seem to use the trimmer during search too, so it may not work for that. [Here is an example in regex101] (https://regex101.com/r/pQMvFL/1) , it works for latin and non-latin character languages. Have just implemented this in an Angular 14 site to clean the start and end of the search term before executing search. I hardcoded it into the lunr.trimmerSupport.generateTrimmer function and ran the tests and all of them seem to pass, so that's a sign it will work.

@MihaiValentin I can put this into a PR if you like, but obviously being ES6 it is probably not as backwards compatible as what is currently there

[dhdaines]

Yes, a lot of the trimmers have this problem (French one too).

You can sometimes get away with just replacing the language specific one with the default one but as noted above \w in JavaScript doesn't mean the same thing as it does in other regex implementations, and the trimmer isn't included in the search pipeline (see https://github.com/olivernn/lunr.js/issues/532) so you may encounter unexpected behaviour (and low recall) with words beginning and ending with non-ASCII characters

[dhdaines]

In #115 this is fixed in a more systematic way than mentioned above, by using the Unicode definitions, so try that out if you like.