Not sure if I’m going to be any help, but I did something similar recently. I didn’t have to match based on a specific item in a list but by the length. The match specifications used by :ets.select and others can get confusing so I just used the fun2ms in iex to create that.
Now on to your problem. I used :ets.first and :ets.next to iterate thru the table and get every set of value stores. It was an unordered set but it still reaches all the nodes when not operated on concurrently. It does take a few seconds for roughly 170k entries. After which point I just used a case statement and guards on every entry to find what I needed within the lists stored at each table entry. I just did this earlier today so it could probably be optimized further but the amount of time it takes isn’t too long for what I’m doing. If you need something in a few milliseconds, I’d go with a different approach.
I understand your remark, but actually my initial question is definitely part of the problem. If i want to extend the elixir school example, I’m an intermediary recruiter, as so I have a amount of developer profiles with some attributes describing their abilities (eg known languages, country, city etc). When i receive a job offer, my job is to match a set of dev on this offer.
Even if it s a multicirteria match, the main question I had (as a newbie I repeat i m not by any mean a functional programer yet :), wanting to test something more evolved that a regular list of map/struct, was how to filter on this list stored in a map in ETS. I deliberately reduced the scope because adding “andalso” clause didn’t seem too complicated.
So considering only this scope, everyone answered fully my question (and more actually I understood the complexity limitation chosen on guard clause, had some alternative proposal, some real world experiment). For now, being still in the learning slope, I just try to address the functional problem (multicriteria match) before trying to optimize anything. My remark on benchmark is just a rule of thumb concerning optimization investigation, I m not at this stage now so this post won’t flow on this side of the topic
In term of volume, I will deal with less (10x - 15 less actually) so i will definitely give a try to the foldl solution given by @peerreynders (which should have the same complexity as yours). Even if it does not fit, i will learn from this option.
