Issue #16989 has been updated by marcandre (Marc-Andre Lafortune).


matz (Yukihiro Matsumoto) wrote in #note-17:
> I agree with some of your proposals (#16990, #16991, #16993, #16995). I want @knu to work on this. If I missed something, he will tell us.

Thank you for reviewing these :-)

> I strongly disagree with #16994. There's no evidence we need frozen sets of strings or symbols that much. Even if we do, I think frozen arrays should come first.

Right, I agree that there is a larger discussion about frozen & static literals to be had.

Would a non-frozen notation for Set of strings / symbols have a chance of being accepted? I would like to avoid `Set.new(%w[a list of words])` as can be currently seen in `RuboCop` or in Rails:

https://github.com/rails/rails/blob/master/actionpack/lib/action_dispatch/http/cache.rb#L128
https://github.com/rails/rails/blob/master/actionpack/lib/action_dispatch/http/headers.rb#L25-L44

I think that many gems have simply not really taken care about `Sets`. For example, this file in `Bundler` defines a 40-element array constant, only to call `include?` on it...

https://github.com/rubygems/rubygems/blob/master/bundler/lib/bundler/settings.rb#L9

I feel that a builtin way to write this might help.

----------------------------------------
Feature #16989: Sets: need 
https://bugs.ruby-lang.org/issues/16989#change-87434

* Author: marcandre (Marc-Andre Lafortune)
* Status: Assigned
* Priority: Normal
* Assignee: knu (Akinori MUSHA)
----------------------------------------
I am opening a series of feature requests on `Set`, all of them based on this usecase.

The main usecase I have in mind is my recent experience with `RuboCop`. I noticed a big number of frozen arrays being used only to later call `include?` on them. This is `O(n)` instead of `O(1)`.

Trying to convert them to `Set`s causes major compatibility issues, as well as very frustrating situations and some cases that would make them much less efficient.

Because of these incompatibilities, `RuboCop` is in the process of using a custom class based on `Array` with optimized `include?` and `===`. `RuboCop` runs multiple checks on Ruby code. Those checks are called cops. `RuboCop` performance is (IMO) pretty bad and some cops  currently are in `O(n^2)` where n is the size of the code being inspected. Even given these extremely inefficient cops, optimizing the 100+ such arrays (most of which are quite small btw) gave a 5% speed boost.

RuboCop PRs for reference: https://github.com/rubocop-hq/rubocop-ast/pull/29
https://github.com/rubocop-hq/rubocop/pull/8133

My experience tells me that there are many other opportunities to use `Set`s that are missed because `Set`s are not builtin, not known enough and have no shorthand notation.

In this issue I'd like to concentrate the discussion on the following request: `Set`s should be core objects, in the same way that `Complex` were not and are now. Some of the upcoming feature requests would be easier (or only possible) to implement were `Set`s builtin.



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:ruby-core-request / ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>