Abstracting the Geocoder from Lucene

[Drabble]

The current implementation of the Geocoder is dependant on Lucene.

This has benefits because it is very easy to insantiate a custom Geocoder inline like so :

Geocoder geocoder =
        new Geocoder(path) {
          @Override
          protected Analyzer analyzer() throws IOException {
            return new StandardAnalyzer();
          }

          @Override
          protected Stream<Document> documents() throws IOException {
            Document d1 = new Document();
            d1.add(new Field("value", v1, TextField.TYPE_STORED));
            Document d2 = new Document();
            d2.add(new Field("value", v2, TextField.TYPE_STORED));
            return Stream.of(d1, d2);
          }

          @Override
          protected Query query(Analyzer analyzer, Request request) throws ParseException {
            return new QueryParser("value", analyzer).parse(request.query());
          }
        };

However if we'd like to make another Geocoder that is not dependant on Lucene, maybe one that uses a different search engine, we are stuck.

The Geocoder also returns a Response object that contains Lucene specific documents which should be abstracted too.

[bchapuis]

I agree, we will probably have to provide an abstraction to hide the TopDocs, ScoreDoc and Document classes to the user. I suggest that we focus on having a good Lucene-based implementation of the Geocoder for now and keep this aspect in mind once we have a use case for geocoder using another search technology (e.g. Postgres, Sqlite, etc.).