Max Inden

> Ensuring that we only pass a slice of at most 65535 bytes if the kernel supports GSO. If I recall correctly, the maximum UDP datagram size (here 65535 bytes)...

Might be resolved through https://github.com/quinn-rs/quinn/pull/2010 which fixes the invalid `cfg`. I introduced this in https://github.com/quinn-rs/quinn/pull/1915. Sorry about that.

Not sure what the cause is here. I too have been debugging `EINVAL` failures, e.g. [due to Android < 26 not supporting IP TOS](https://github.com/quinn-rs/quinn/pull/2079), in the past. While another datapoint...

Using `SystemTime` sounds like the reasonable choice to me. I don't understand the benefit of nesting a `SystemTime` in an additional `enum`. Can you expand @recatek?

@larseggert this pull request should resolve all issues raised in https://github.com/quinn-rs/quinn/pull/1993#pullrequestreview-2346382573. Can you take a look? Targeting `main` directly given that changes are not tied to https://github.com/quinn-rs/quinn/pull/1993.

@djc @Ralith ready for a full review. Let me know if you find this pull request helpful. Will unblock https://github.com/quinn-rs/quinn/pull/1993.

Thank you for the review @djc. I applied all suggestions. In addition, I split the one large commit into smaller atomic commits. Let me know what you think.

Windows CI failure seems intermittent. See e.g. https://github.com/mxinden/quinn/actions/runs/11424992884/job/31786093629 for a succeeding run, simply adding a `RUST_BACKTRACE=1`, and a recent failure https://github.com/quinn-rs/quinn/actions/runs/11478020814/job/31941343750?pr=2010.

Thank you for the review @Ralith. Would you mind taking another look?

> A targeted quinn-udp benchmark might be more useful. How about using the `throughput.rs` benchmark @larseggert? https://github.com/quinn-rs/quinn/blob/main/quinn-udp/benches/throughput.rs