0.10 release planning about quinn

[djc]

Draft changelog:

• Initial support for PLPMTUD (#1510 with followup in #1529, thanks to aochagavia)
• Bump rustls to 0.21.0 (#1515, thanks to cpu)
• Fix: build and tests on FreeBSD 13.2 (#1555, thanks to Tirka)
• Fail FreeBSD CI if any step fails (#1557, thanks to Tirka)
• Fragmentation handling improvements (#1547)
• Add missing exports of stats types (#1546)
• Rename min_guaranteed_mtu to min_mtu for clarity (#1552)
• Duplicate Transmit and EcnCodepoint across -proto and -udp (#1541)
• quinn: take Arc<dyn Runtime> directly (#1534)
• Support external crypto implementations (#1496, thanks to nmittler)
• Do not require &mut self in AsyncUdpSocket::poll_send (#1519, thanks to dignifiedquire)
• Getters for stream concurrency metrics and related documentation (#1518, thanks to FlorianUekermann)
• Fallback to not using ECN if IP_TOS is not supported (#1516, thanks to link2xt)
• Implement fallback for sendmmsg and recvmmsg (#1504, thanks to link2xt)
• Borrow self in read_to_end, rather than consuming (#1502)
• Improve error when connecting with an unsupported version (#1482)
• Fix racy Endpoint::wait_idle (#1477)
• Re-export udp::{RecvMeta, UdpState} for AsyncUdpSocket (#1476, thanks to XOR-op)
• Route logs via log crate by default when tracing isn't used (#1473)
• Fail gracefully on attempts to connect to an unspecified IP address (#1474)
• add FreeBSD IP_RECVDSTADDR support (#1447, thanks to lucifer9)
• udp: add safe wrapper for setsockopt() (#1461)

Documentation

• Allow initial MTU to be specified in bulk benchmark (#1456)
• Document how to close a stream correctly (#1493, thanks to flub)
• Update wording about QUIC being a draft (#1468, thanks to zicklag)
• Clarify async runtime support (#1497, thanks to mxinden)
• Fix usage message (#1484, thanks to dzvon)

Internal improvements

• Allow benchmarking without packet protection (#1436, thanks to stormshield-damiend)
• Fix no-protection option in perf crate (#1550, thanks to aochagavia)
• Don't rely on sleeping in 0-RTT test (#1522)
• Update deny.toml (#1470, thanks to NukeManDan)
• Bump quinn-proto version (#1520, thanks to nmittler)
• Clippy fixes (#1466, thanks to StygianLightning)
• Bump MSRV to 1.63 (#1548)
• Bump quinn-udp version to 0.4 (#1535, thanks to link2xt)

[djc]

It's done! https://github.com/quinn-rs/quinn/releases/tag/0.10.0

[mxinden]

Should we try to get #1219 in?

That would be wonderful! 🙏

[mxinden]

How do we plan for QUIC V2?

IIRC QUIC V2 is mainly an anti-ossification measure that won't impact public API. It'd be healthy for us to have it but I don't think it's pressing. Might bear a double-check.

Correct, see:

This document specifies QUIC version 2, which is identical to QUIC version 1 except for some trivial details. Its purpose is to combat various ossification vectors and exercise the version negotiation framework. It also serves as a template for the minimum changes in any future version of QUIC.

Note that "version 2" is an informal name for this proposal that indicates it is the second standards-track QUIC version. The protocol specified here uses a version number other than 2 in the wire image, in order to minimize ossification risk.

https://datatracker.ietf.org/doc/draft-ietf-quic-v2/

[abonander]

Should we consider bumping the next version to 1.0?

As long as rustls is part of the public API, I wouldn't recommend that.

[djc]

We should probably get #1529 in before release. I think that's probably the only thing blocking the release at this point?

[djc]

We should review #1545 and #1547 before we decide to release. I think those will be relatively straightforward, though.

[Ralith]

I'd like to include mechanics for enabling MTUD automatically where supported (some sort of PlatformConfig?), since most users won't and shouldn't need to bother with target-specific configuration.

How do we plan for QUIC V2?

IIRC QUIC V2 is mainly an anti-ossification measure that won't impact public API. It'd be healthy for us to have it but I don't think it's pressing. Might bear a double-check.

Should we consider bumping the next version to 1.0?

I agree we're getting close, but if we're going to treat 1.0 as special we should probably defer a decision until we've actually not broken for a good while. Alternatively, we could abandon cosmetic concerns and just start bumping the major version without a care.

[Ralith]

Should we try to get #1219 in?

As much as I'd love to get this merged finally, it doesn't need to be a blocker since it's not API-breaking, and can easily be included in a patch release. If it continues to be a difficult review it probably shouldn't delay the other more practically beneficial changes we have staged.

[Ralith]

I think so; I should be able to review that within the next few days.

[XOR-op]

Since #1529 has been merged, should we plan to release 0.10? It looks like all the checkpoints have been resolved.

[Ralith]

Rename min_guaranteed_mtu to min_mtu for clarity

This is a bit confusing as a literal changelog entry because in 0.9 the parameter was called initial_max_udp_payload_size.

udp: add safe wrapper for setsockopt()

This is purely internal.

LGTM otherwise!