feat: Add some benches and integrate with CodSpeed

[larseggert]

No description provided.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.03%. Comparing base (2b341bf) to head (4bcf894).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #62   +/-   ##
=======================================
  Coverage   97.03%   97.03%           
=======================================
  Files           2        2           
  Lines         776      776           
  Branches      776      776           
=======================================
  Hits          753      753           
  Misses         21       21           
  Partials        2        2           

Flag	Coverage Δ
linux	97.03% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[codspeed-hq]

Congrats! CodSpeed is installed 🎉

🆕 8 new benchmarks were detected.

You will start to see performance impacts in the reports once the benchmarks are run from your default branch.

Detected benchmarks

• simple_ipv6_only (Simulation): 19 µs
• many_endpoints (Simulation): 149.5 µs
• dual_stack_racing (Simulation): 25.1 µs
• https_with_ech (Simulation): 23.4 µs
• many_endpoints (Memory): 4.1 KB
• https_with_ech (Memory): 1.2 KB
• dual_stack_racing (Memory): 1.4 KB
• simple_ipv6_only (Memory): 1 KB

---

[Copilot]

Pull request overview

Adds a benchmark suite for the happy-eyeballs state machine and wires it into CI via CodSpeed to track performance over time.

Changes:

• Added divan-style benchmarks covering several Happy Eyeballs scenarios (IPv6-only, dual-stack racing, HTTPS/ECH, many endpoints).
• Integrated CodSpeed-compatible Divan dependency and registered the happy_eyeballs bench target.
• Updated CI to run CodSpeed benchmarks and expanded coverage command to include all targets.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 1 comment.

File	Description
benches/happy_eyeballs.rs	New Divan/CodSpeed-compatible benchmark scenarios for the state machine.
Cargo.toml	Adds CodSpeed Divan compat dev-dependency and declares the bench target (no harness).
Cargo.lock	Locks new transitive dependencies introduced by CodSpeed/Divan compat.
.github/workflows/rust.yml	Runs cargo bench --bench happy_eyeballs via CodSpeed; adjusts coverage run to --all-targets.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[larseggert]

@mxinden ready from my side. Can tweak the actual benches in follow-up PRs.

[github-actions]

Good addition — the benchmarks are well-structured, cover the key state machine paths (single-family, dual-stack racing, HTTPS/ECH, many-endpoint exhaustion), and the CodSpeed integration is straightforward.

A few things to address:

• Action version mismatch: The benchmarks job still pins mozilla/actions/rust at v1.0.4 while the rest of the workflow was bumped to v1.0.11 via main. See inline comment.

• --all-targets in coverage job: This change causes the benchmark binary to also execute during coverage runs (divan falls back to walltime mode outside CodSpeed). Worth considering whether that's intentional — see inline comment.

• PR title typo: "Cosdpeed" → "CodSpeed".

• Re: Copilot's None-in-loop finding (comment): This was already addressed — the current code correctly groups None with the break arm (lines 253–256).

• Re: zizmor permissions finding (comment): Already addressed — id-token: write has the explanatory comment on line 50.

[larseggert]

@mxinden I don't think this will cause vendoring issues for Gecko.

[larseggert]

@mxinden OK to merge?

[github-actions]

Good set of benchmarks — they cover the key state machine paths (single-family, dual-stack racing, HTTPS/ECH, many-endpoint exhaustion) and the CodSpeed integration is clean.

Most findings from prior reviews have been addressed in the latest push: --all-targets removed from coverage (replaced with cargo check --benches), String::new() for failure messages, trailing newline in Cargo.toml, and None correctly breaks in the many_endpoints loop.

One remaining item: the mozilla/actions/rust pin in the benchmarks job is still behind — v1.0.11 vs v1.1.0 in the rest of the workflow (see inline comment).

PR title typo (noted by the prior review but still present): "Cosdpeed" → "CodSpeed".

[github-actions]

Well-designed benchmark suite covering the key state machine paths (single-family, dual-stack racing, HTTPS/ECH, many-endpoint exhaustion). The CodSpeed integration and CI wiring are clean.

Blocking: The latest commit ("Fix") breaks CI — cargo check --benches and cargo codspeed build both fail. See inline comment for the fix.

Previously flagged issues — now resolved:

• mozilla/actions/rust pin: now v1.1.0, consistent with the rest of the workflow
• --all-targets in coverage: replaced with a separate cargo check --benches step
• String::new() for failure messages (avoids allocation noise)
• Trailing newline in Cargo.toml
• None correctly breaks in the many_endpoints loop

[github-actions]

Well-structured benchmark suite covering the key state machine paths — good selection of scenarios (IPv6-only baseline, dual-stack racing, HTTPS/ECH with hints, many-endpoint exhaustion).

Status of prior review findings — all addressed in latest push:

• assert_matches import (prior comment): Fixed — switched to stable assert!(matches!(…)), removing the broken use std::assert_matches that prevented compilation.
• mozilla/actions/rust pin: Now v1.1.0, consistent with the rest of the workflow.
• --all-targets in coverage: Replaced with a dedicated cargo check --benches step — cleaner separation of compile-check vs. coverage.
• String::new() for failure messages: Eliminates per-iteration allocation noise.
• Trailing newline in Cargo.toml: Present.

No new blocking issues found. A few observations:

• The many_endpoints loop driver (lines 228–241) is well-designed: it correctly breaks on terminal outputs (Succeeded/Failed) and None, and the VecDeque-based in-flight tracking naturally handles the interleaved AttemptConnection/Timer cadence the state machine produces.
• The code duplication between benches/happy_eyeballs.rs and tests/common/mod.rs (constants, input constructors) was noted in a prior review — still present but non-blocking and the bench helpers are intentionally leaner (no env_logger, drain_dns_queries baked into setup).
• CI permissions are properly scoped (id-token: write only on the benchmarks job, with an explanatory comment). The workflow_dispatch trigger for CodSpeed backtesting is standard.

[mxinden]

Sorry for the delay.

The state machine is quite simple. The most expensive operations are a couple of string allocations.

I'm not convinced these benchmarks provide much value at this point. If the result is X, what does X tell us relative to? We haven't identified any improvement opportunities to measure against.

I'd suggest holding off on merging until we've identified something concrete to improve, e.g. if happy eyeballs shows up in a Firefox profile. Until then, I suggest we keep this as a pull request only. I should have suggested this earlier, sorry about that.

[larseggert]

ACK, I'll make it a draft.

[larseggert]

(Wondering if the memory benchmarks would be more useful...)