Paul Loyd

Thanks for the quick response, It sounds reasonable. This behavior should be documented anyway. Initially, I tried to implement "detachable" streams that can be added later to the same `StreamsUnordered`...

I think the crate should provide ABA protection. `Header` already contains `index` which can also include the generation number. It makes `get(_mut)` and potential `remove()` safer and reduces sporadic wakeups...

Although, I understand that the main reason for the benchmark is to make loud statements: > Column-oriented design that operates directly with blocks of data > [Dramatically more efficient](https://github.com/ClickHouse/ch-bench) >...

Linked PR is activity

The bot's logic is ill-conceived =(. It should be reopened

The documentation is not precise enough. It *literally* means "serialize as `Array((K, V))`", so you can use `Vec` and so on here. What's `Map`, `serialize_map`, and (most important!) `Value` in...

It's possible to support real maps (`HashMap`, `BTreeMap`, `IndexMap` etc) directly without using `&[tuple]`, but it opens a large space for bugs because of `#[serde(flatten)]`, `serde_value::Value` and other things which...

So, it's already implemented, am I right?

Actually, it's already implemented in `unstable-stuck-detection` feature.

What kind of deadlock do you see here? I'm always planning to handle responses unboundedly (so `ctx.respond(token)` is non-blocking already), so I don't see where deadlock can occur for now....