Can we have an API similar to `HashMap::retain`?

[linrongbin16]

Hi, can we have an API similar to HashMap::retain:

pub fn retain<F>(&mut self, f: F)
where
    F: FnMut(&K, &mut V) -> bool

It only retains all the elements that satisfies the f predicate, and removes all other elements?

Actually I don't mind to use the drain API to archive the same result, I think I could first use drain to empty the cache, and iterate all the elements and insert back those satisfy the f predicate.

Do they have the same performance?

[arthurprs]

It's possible, yes. Would you like to submit a PR for it?

retain would probably be better, as the retained elements would retain their cold/hot/frequency information. If you drain and re-insert the entries, they would end up with random-ish cold/hot/frequency information.

[linrongbin16]

I can have a try :)

[linrongbin16]

hi @arthurprs , I plan to implement the retain API (for both sync and unsync cache) like this:

1. I think I will have to add a low-level api retain in the CacheShard struct, then wrap it in both sync::Cache and unsync::Cache.
2. I will first iterate the map inside the CacheShard, so I will have the idx token to get the Entry::Resident(r) (and the key and value), calculate the predicate f result, and get the filtered collection, that all the elements have the false result (they should be removed from the cache).
3. Iterate the collection again, and use the remove_internal api to actually remove the element.
   Can you give some comments about this solution?

Also I have some questions:

1. If I use this solution, is it thread safe for the sync::Cache? I mean, I won't add any Mutex or Lock to keep the data safe in multiple threadings.
2. Do we need to make the Val in the predicate f API mutable? I mean, do we want the f: F signature to be F: FnMut(&Key, &mut Val) -> bool, or just FnMut(&Key, &Val) -> bool

---

Here's the PR: #64

---

Update: I think the PR is ready for review, @arthurprs would you please take a look?