Vitali Lovich

I wonder if there's a more direct repro case we can construct...

Hmm.... not sure I'm seeing the same thing. What I did was I moved that `blocking_thread` construction as a standalone line and then added surrounding `println`: ``` + println!("starting blocking...

Actually, nothing even to do with MeshIPC / LocalExecutorPoolBuilder necessarily I think. Here's an even simpler scenario that also seems to hang https://gist.github.com/vlovich/6876632b48df4289eb3e05716f9f431a - only uses tokio sync and a...

It's something to do with the Context / Waker but I thought that was a Rust thing? After instrumenting tokio oneshot internals, I see `wake_by_ref` being called on the send...

Well, the latest repro disappears if I use `--fair-sched=yes`. A deadlock on the stderr futex within the Rust stdlib reappears (even with `--fair-sched=yes`) if I add a bunch of logging...

I also noticed problems with flume channels not processing for the blocking operations (e.g. copy_file_range) under valgrind. If I add an `eprintln` or `std::thread::sleep(std::time::Duration::from_millis(1));` after `tx.send_async` in `BlockingThreadPool::push` then the...

If I'm reading it correctly, it sounds like like what @seanmonstar is actually saying is that `#[non_exhaustive]` just needs to be an annotation on the ErrorKind or to use the...

As a workaround, feeding the IconSource::Data the raw .ico file without decoding does work. The inconsistency between platforms in the APIs is quite unfortunate. Would be nice if that were...

I thought it was but need to fix failing tests. I actually don't mind the clippy stuff in CI. Keeps the codebase consistent which outweighs the nitpickiness annoyance of it...

BTW the failure here was just a normal compile error :).