Search is implemented with Elasticsearch (es). Data in es is considered volatile and it should always be possible to rebuild the index from other data sources (postgres in this case).
Install Elasticsearch from http://www.elasticsearch.org/overview/elkdownloads/. Install the ICU plugin https://github.com/elasticsearch/elasticsearch-analysis-icu, making sure you select the proper version for your version of es. Once set up, recreate es mappings by invoking the elasticsearch:recreate command and populate the index by invoking elasticsearch:populate (this might take a long time you your subtitle cache is empty).
Edit synonyms.esn file (see http://www.elasticsearch.org/guide/en/elasticsearch/guide/current/synonyms-expand-or-contract.html). Then invoke elasticsearch:build-synonyms which will update synonyms.compiled.neon. Make sure the compiled file is ok. Commit both files.
TODO: actually, the czech file now needs to be called synonyms.compiled.cs.neon
All synonym changes require reindexing.
We use whitespace tokenizer so we can add atp. as stopword while retaining atp as adenosintrifosfát.
In order to remove the special characters we use pattern_replace filter.
To enable and optimize elastic search for a specific language: cp app/config/elastic.en.neon app/config/elastic.<your_locale>.neon
where <your_local> is set in config.local.neon