The point with ets:foldl/3 is that you don’t have to handle the full table scan manually - the ets module does it for you (it’s still a full table scan though). You simply give it the table, an initial value (likely an empty list) and a function fun((Element :: term(), AccIn) -> AccOut). The function is presented with each element in the table in turn and it can then stuff any element “it likes” in the accumulator (e.g. the list). The completed accumulator is then returned to you.

You should be able to test your function with Lists.foldl/3.

If you’re thinking about indexing, you might want to consider Mnesia, because it supports that feature out of the box. As others suggested, you’ll likely need a bag table, with fields user_id (key), and language. Then, you can add a secondary index on the language field.

With :ets you get fast lookups if your key is part of the match.

In the elixir school example, “Java” is not part of the key but is a value. In which case ets has to do a full table scan which means it has to try and match all the rows in it. You are on the right track about reordering data. Inverting your values and keys and storing them in another ets table should give you the best performance. Also know that there is a :bag type of ets table available which can be used in this case.

I had a similar situation where I had to first lookup on a country and then city, which wouldn’t be the right fit performance wise for ets. So, I flipped the order and maintained a bag with {city, country}.

Ok, I understand the constraints on guards now. So basically I should handle a full table scan manually which does not sound that efficient if the storage is filled with few thousands of records. Unless…

… you build an index

Thanks guys to bring some light on my question

Fred

You will need to retrieve the value and test it. Only calls that can be trivially inlined and are O(1) cost (or more specifically, the ones that the BEAM VM can turn into special optimized calls) can be guards. So yeah, just grab and test, or use something like PostgreSQL.