Fixing the naive implementation of sleep #1523
Conversation
|
Instead of creating the atomic during poll, why not create it when creating
the Sleep object?
|
Can do that and improve. Wanted to see if this approach works. I had the Issue opened to hope and have the discussion there but since it didnt move i thought i would just open a PR. Also initial implementation thought was to launch thread on first poll |
|
Can do that and improve. Wanted to see if this approach works. I had the
Issue opened to hope and have the discussion there but since it didnt move
i thought i would just open a PR.
Many (most) of us take much of December off, hence the delay. I'm
responding via email as I'm away from work.
I would also like to split off the tokio part as it's own off-by-default
feature (and PR), and keep this pr to just fixing the bug you're
experiencing.
|
Also out of office but getting to play around with things :). Thanks for replying and reviewing the PR. Will split it in that manner |
|
Made the tokio sleep alternative opt-in disabled by default. The naive implementation does a let of work in the poll as it needs to grab the wake context to wake the blocked task. |
| azurite_workaround = [] | ||
| xml = ["quick-xml"] | ||
| tokio-fs = ["tokio/fs", "tokio/io-util"] | ||
| tokio-fs = ["tokio/fs", "tokio/sync", "tokio/io-util"] |
There was a problem hiding this comment.
What changed to the tokio-fs feature that requires tokio/sync?
There was a problem hiding this comment.
It uses mutex which is behind sync feature flag. It just didn't compile. Should fix ci tests as well
2 participants
Address Issue highlighted in #1515 in the naive implementation.
Delegates to tokio's sleep if tokio feature is turned on