Simplify WaitGroup implementation #958
Conversation
✅ Deploy Preview for salsa-rs canceled.
|
CodSpeed Performance ReportMerging #958 will degrade performances by 5.28%Comparing Summary
Benchmarks breakdown
|
|
Is there any other benefit than code size reduction. It seems codspeed doesn't like it at all. |
7a3884e to
ef1c5dc
Compare
|
No there isn't. I was curious about the perf impact, I agree this isn't great. Checking if we can actually simplify our impl now |
ef1c5dc to
1b15e41
Compare
WaitGroup impl with crossbeam'sWaitGroup implementation
|
Hm, we could get rid of the additional |
|
@ibraheemdev could you take a look at this PR? |
f1f158f to
ee184dd
Compare
Co-authored-by: Ibraheem Ahmed <ibraheem@ibraheem.ca>
ee184dd to
9b4da36
Compare
|
This introduced a potential to deadlock. I'm still investigating how exactly, but my guess is that the atomic ordering is not correct (it's not at least in one thing, but that doesn't fix the deadlock). |
This PR can create deadlocks - there is nothing synchronizing the mutex/condvar with the `Arc` of the `Zalsa`, therefore code may wake up by the `Condvar` but not see an updated refcount, go to sleep again, and never wake up.
No description provided.