Stepan Koltsov

> The merge style is much more helpful when trying to arrive at those sorts of judgements. That's a subjective opinion, and it is wrong because it is presented as...

> I find that cfg-guarded imports are harder to manage with your preferred arrangement Well, it depends. I had a scenario in mind (from various my projects) where they are...

> The only [reason](https://abseil.io/docs/cpp/guides/container#abslnode_hash_map-and-abslnode_hash_set) you would want to use them over a normal hash map is if you need pointer stability for some reason This is actually not the only...

No I didn't do benchmark, sorry. It is not trivial to do testing with unreleased versions in our repo setup. But there are more reasons to do this change. Another...

I reimplemented the library in our project with two PRs I submitted and more changes. This is the implementation. https://gist.github.com/stepancheg/9d2ebac23b27ff8d21a1bcf494b5ac7c The speedup in our program (not just internment) is about...

`raw_entry_mut` should probably work. I didn't know such function exists.

> I'll think about how to store a RandomState outside a Mutex `static RANDOM_STATE: OnceCell = ...` should not have noticeable performance overhead, no?

Assigning the PR to myself. However, this week we have internal conference, and next week I'm on holidays. Can it wait two weeks @aherrmann?

@aherrmann working on it. Sorry for delay!

I'm working on some changes in `dynamic_output` API which I plan to land before taking this PR. I'll rebase this PR on top of my work. Just FYI, this PR...