fitzgen / associative-cache Goto Github PK

When running tests for 1.0.1 on i386, I get the following failures:

$ /usr/bin/cargo test -j12 -Z avoid-dev-deps --release --no-fail-fast --lib
running 15 tests
test replacement::lru::tests::lru_replacement ... ok
test indices::tests::pointer_two_way ... FAILED
test indices::tests::pointer_direct_mapped ... FAILED
test tests::entry ... ok
test tests::get ... ok
test tests::get_mut ... ok
test tests::capacity ... ok
test tests::insert ... ok
test tests::iter ... ok
test tests::into_iter ... ok
test tests::iter_mut ... ok
test tests::len ... ok
test tests::replacement_policy ... ok
test tests::retain ... ok
test tests::remove ... ok

failures:

---- indices::tests::pointer_two_way stdout ----
thread 'indices::tests::pointer_two_way' panicked at 'assertion failed: `(left == right)`
  left: `0..2`,
 right: `2..4`', src/indices.rs:185:9

---- indices::tests::pointer_direct_mapped stdout ----
thread 'indices::tests::pointer_direct_mapped' panicked at 'assertion failed: `(left == right)`
  left: `2..3`,
 right: `1..2`', src/indices.rs:161:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    indices::tests::pointer_direct_mapped
    indices::tests::pointer_two_way

test result: FAILED. 13 passed; 2 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

error: test failed, to rerun pass `--lib`
error: 1 target failed:
    `--lib`

Hi, I was looking for something exactly like your crate for a class project. The only thing that I was wondering was the (amortized) time complexity of get, get_mut, insert, and remove? Is there significant overhead to using this associative cache over a regular hash map?

This crate claims that its license is MIT OR Apache-2.0, but the license text is not included. Both the MIT and the Apache-2.0 licenses require including the full text when distributing the software, so users cannot comply with that license if you don't include it. Please add the license text.

context:

pub trait LruTimestamp {
    type Timestamp: PartialOrd;
    fn get_timestamp(&self) -> &Self::Timestamp;
}

I think this means LruTimestamp::get_timestamp() has to return a reference to something inside self, so it can't compute the timestamp. For example, if I want the timestamp to be self.member1 + self.member2. Or in my case, self is #[repr(packed)], so I can't[*] return a reference to a member; I want to do the unaligned dereference and return the value.

[*] or at least, rustc says it will be disallowed in a future release

I think that the interface could be changed to:

pub trait LruTimestamp {
    type Timestamp: PartialOrd;
    fn get_timestamp(&self) -> Self::Timestamp; // removed &
}

And then you can still return a reference to something inside self when desired, by defining the Timestamp type to be a reference, e.g.:

impl<T> LruTimestamp for WithLruTimestamp<T> {
    type Timestamp = &Cell<Instant>; // added &
    fn get_timestamp(&self) -> Self::Timestamp { // removed &
        &self.timestamp
    }
}

I'm not sure exactly how to do this as a non-breaking change, maybe add a new trait, LruTimestampByValue?