Benjamin Saunders

> After app come to foreground socket become writable again, but probably estimation of available bandwidth was "ruined" The new interface proposed in #2018 might help accelerate recovery, though it...

Yeah. Luckily it's a clean backport, so risk should be pretty low. Just need a decision on the changed default.

It's been long enough that the downstream I care about has since updated to 1.x anyway.

Calling trait *methods* in `const` isn't, so I'm not sure you'll be able to constify any useful constructors.

Another interesting variable is memory use. Each connection has a send buffer that may grow as required to fit a GSO batch. If e.g. a Windows server, capable of 512-datagram...

macOS doesn't actually support segmentation offload, so that works fine. Thanks!

Interesting problem, and one I expect to face eventually as well. My instinct is that this would be best solved by a separate, specialized `Signal` which automatically ramps volume to...

I'm leery of schemes that rely on juggling high-level `Signal`s because, by design, those operations are not real-time. You can mitigate this (and perhaps this is your intention) by arranging...

> Only the underrun "coda" (cycle + fade out) is pre-queued Right; I'm not convinced that constantly re-queueing underrun signals that we rarely actually reach is the way to go....

Something like that, yeah. A good timestamp format is a u64 sample number. Maybe that's what the API should work in terms of. > Probably choosing a time origin at...