Another LRU crate about lru-rs

[jeromefroe]

Hi @marmeladema! It's no bother at all!

I just took a look at your crate, and I think it's a great approach! I'd be happy to add a comment to the README linking to it and explaining the differences between the two. The way I see it, the two crates occupy slightly different positions on the spectrum of performance and usability. clru-rs offers an API closer to HashMap and less unsafe code at the cost of slightly more memory usage because the keys are stored in a vector instead of a linked list, and this crate makes the opposite trade-off. Does that sound about right?

[marmeladema]

By the way, I haven't tagged a new release yet, but clru does not pre-allocate unconditionally all the memory anymore.
There is still a constructor available to do so though, for cases (like the ones I have) where it can be known in advance that the cache is going to be full.

[marmeladema]

Thank you for taking the time to look at the code! Really appreciate it!

I'd be happy to add a comment to the README linking to it and explaining the differences between the two.

Wow that would be super nice 👍 Maybe it's a bit too soon though, I'd like some more people to look at it / use it.

at the cost of slightly more memory usage because the keys are stored in a vector instead of a linked list

The keys are moved in a Rc and thus are shallow copy in the linked list. The overhead with this would be similar as if the keys were boxed once. I am not sure there is such a difference in memory usage between the two crates? It would be nice to measure it though! The only thing that I can think of where clru might be more memory hungry would be for non-full cache, because it pre-allocates all nodes ahead of time but then there is not much point in using an LRU cache that is never full.

[jeromefroe]

Maybe it's a bit too soon though, I'd like some more people to look at it / use it.

Sounds good. We can circle back whenever you feel it's ready!

The only thing that I can think of where clru might be more memory hungry would be for non-full cache, because it pre-allocates all nodes ahead of time but then there is not much point in using an LRU cache that is never full.

That was all I could think of too. And I agree, it certainly seems like the difference would be negligible.