Kiran Pamnany

I have no idea how I missed this notification, very sorry. So, if a task `wait`s on a `Condition`, that involves two steps: 1) adding itself to a list of...

Typically, condition variables are implemented such that the two steps I mention above (add to queue and yield) happen atomically, to prevent lost wakeups. We dodge this problem with our...

I instrumented Julia to measure the time lost due to these discards. For the fib code in the issue, I see a bit over 1ms lost with 8 threads. Certainly...

I tested this instrumentation internally and found that the cost of this scheduler quirk is roughly 0.2% for most of our code, with a maximum of 3% for one outlier....

From the backtrace, it looks to me like a search through module bindings segfaulted, which is different from https://github.com/JuliaLang/julia/issues/51267. I was wondering if https://github.com/JuliaLang/julia/pull/49556 perhaps unintentionally broke something and was...

> I think you need to keep this mostly inside flisp (particularly parsed_method_stack). I also don't think the module name has much value in being mangled into it. Making the...

We've tested this change by exercising it through compilation warmup in our application. Without this, we saw roughly 53% warmup effectiveness when upgrading versions. With this, we're seeing 68% warmup...

Ran more tests to verify the gains from this and have verified that this is a positive change for us. Any objections to merging @JeffBezanson?

Also +1 to Kristoffer's comment. Edit: with #510, I can make this work for me, but still think decoupling `julia` and `juliaup` environment variables is a better choice.

Attn: @vchuravy and @Keno