Add a --detach flag to bare metal bencher run

[epompeii]

This changeset adds a --detach flag to the bare metal bencher run command. This allows you to skip awaiting the bare metal job execution synchronously.

[github-actions]

🤖 Claude Code Review

PR: #762
Base: devel
Head: u/ep/detach
Commit: 7d6d6849b779ae3564551b6c90dda56bc986c47d

---

Here's my review:

---

PR Review: --detach flag for bencher run + human display improvements

Summary

Adds a --detach flag to bencher run for fire-and-forget remote job submission, renames --poll-interval to --job-poll-interval (with backward-compat alias), improves human-readable report output, and adds a detach integration test.

Issues

1. poll_interval field is populated but unused when detach: true

services/cli/src/bencher/sub/run/mod.rs:131 — When detach is true, the poll_interval field is still parsed and stored in the Job struct, but never used. This is harmless but slightly misleading. The conflicts_with on the clap side handles this correctly, so it's fine in practice.

2. Test uses wall-clock time with std::time::Instant and std::thread::sleep

tasks/test_api/src/task/plus/runner.rs:478-485 — The detach test polls with real wall-clock time (Instant::now(), thread::sleep). Per CLAUDE.md, time-based tests should be deterministic. However, this is an integration/smoke test in tasks/ (not a unit test), and it's polling an external process, so real time is appropriate here. This is acceptable.

3. Missing --format json assertion on detach output

tasks/test_api/src/task/plus/runner.rs:455 — The test passes --format json and --quiet, then deserializes stdout as JsonReport. If detach mode ever prints additional non-JSON output to stdout (the cli_eprintln_quietable! currently goes to stderr, which is correct), the test would catch it. Good.

4. Detach test error propagation ordering

tasks/test_api/src/task/plus/runner.rs:250-252 — The error propagation is result? ; no_sandbox_result? ; detach_result?. If an earlier test fails, the detach test is skipped (set to Ok(())). This is intentional to avoid running dependent tests after a failure, which is good.

5. Human output: leading newlines on first content

lib/bencher_comment/src/lib.rs:77-79 — human_report_link writes "View report: {url}" without a leading newline, then human_no_benchmarks and human_results_list start with "\n\n". This produces clean output. Good.

6. UTM stripping for human URLs

lib/bencher_comment/src/lib.rs:677-706 — The resource_url_human method strips UTM params from URLs shown to humans. Clean separation. The resource_url_inner refactor with the utm: bool parameter is straightforward.

Minor Nits

• Commit messages: claude_1 is not descriptive. Consider squashing/rewording before merge.
• TODO remove in due time at services/cli/src/parser/run.rs:274 — Consider adding a version target for removing the --poll-interval alias (e.g., // TODO(v0.7): remove deprecated alias), so it doesn't linger indefinitely.
• The detach_result variable in tasks/test_api/src/task/plus/runner.rs:234 is not actually async or deferred — it runs synchronously inline. The naming pattern matches the surrounding code, so this is fine for consistency.

Looks Good

• Backward compatibility via alias = "poll-interval" for the rename
• conflicts_with = "job_poll_interval" prevents nonsensical --detach --job-poll-interval combos
• Docs added for all 9 languages
• Changelog updated
• Test coverage for the new detach flow
• No security concerns

Verdict

Approve with minor suggestions. The code is clean, well-tested, and follows project conventions. The only actionable item is adding a version target to the deprecation TODO.

---

_{Model: claude-opus-4-6}

[github-actions]

Bencher Report

Branch	u/ep/detach
Testbed	intel-v1

Click to view all benchmark results

Benchmark	Latency	Benchmark Result microseconds (µs) (Result Δ%)	Upper Boundary microseconds (µs) (Limit %)
Adapter::Json	📈 view plot 🚷 view threshold	4.54 µs (-0.07%) Baseline: 4.54 µs	4.64 µs (97.85%)
Adapter::Magic (JSON)	📈 view plot 🚷 view threshold	4.47 µs (+1.24%) Baseline: 4.42 µs	4.47 µs (99.94%)
Adapter::Magic (Rust)	📈 view plot 🚷 view threshold	25.34 µs (+0.91%) Baseline: 25.12 µs	26.61 µs (95.23%)
Adapter::Rust	📈 view plot 🚷 view threshold	3.38 µs (+0.12%) Baseline: 3.37 µs	3.47 µs (97.37%)
Adapter::RustBench	📈 view plot 🚷 view threshold	3.37 µs (+0.01%) Baseline: 3.37 µs	3.49 µs (96.67%)

🐰 View full continuous benchmarking report in Bencher