fix(cli): render sub-millisecond timer inputs correctly in dora node info

[phil-opp]

Issue

build_input_info in binaries/cli/src/command/node/info.rs reconstructed every timer input source by hand as dora/timer/millis/{interval.as_millis()}.

Duration::as_millis() floors to 0 for any sub-millisecond interval, so a valid high-rate timer input such as dora/timer/hz/2000 (500 µs) was displayed by dora node info as dora/timer/millis/0 — a nonsensical zero-interval source that no longer round-trips through the descriptor parser. Whole-Hz timers like dora/timer/hz/3 (~333 ms) were also rendered lossily as millis/333.

Fix

Use InputMapping's canonical Display impl (libraries/message/src/config.rs) instead of re-deriving the source string. Display already selects the coarsest exact unit (secs/millis/micros/nanos) and is the form the descriptor round-trips through. The User and Logs arms of the old hand-rolled match rendered identically to Display anyway, so this also removes a small reimplementation of the trait — build_input_info now just calls input.mapping.to_string().

Validation

• New unit tests were unnecessary: the sub-millisecond round-trip is already covered by dora-message's existing timer_subms_interval_roundtrips test, which guarantees Display never truncates.
• cargo test -p dora-cli, cargo clippy -p dora-cli --all-targets -- -D warnings, and cargo fmt --all -- --check all pass.
• Full workspace test suite green except for pre-existing zenoh/ros2 networking tests that require IPv6 (unavailable in the CI sandbox) — unrelated to this CLI-only change.

---

🤖 This is a machine-generated pull request opened by Claude Code. A human maintainer should review before merging.

🤖 Generated with Claude Code

https://claude.ai/code/session_0142J7B2578ixrh77kNDNJry

---

Generated by Claude Code

[trunk-io]

😎 Merged successfully - details.

[phil-opp]

Automated review run summary

This PR is one of a batch opened by an automated Claude Code review pass over origin/main (84709f7). The pass reviewed the mavlink2-bridge, arrow-convert, message, core, node/operator, daemon, coordinator, telemetry, and CLI subsystems. Each finding below was verified by hand against current main before acting. All PRs are machine-generated drafts — a human maintainer should review before merging.

Opened PRs

Correctness

• fix(cli): render sub-millisecond timer inputs correctly in dora node info #2859 — dora node info rendered sub-millisecond timer inputs (e.g. dora/timer/hz/2000) as dora/timer/millis/0 due to Duration::as_millis() truncation.
• fix(cli): don't drop compact JSONL log lines that omit the level key #2860 — parse_jsonl_line silently dropped compact JSONL log lines missing a level key (a ? short-circuit that also made a .unwrap_or("stdout") default dead code).

Robustness / security

• fix(cli): validate cluster machine id and labels before they reach the remote shell #2861 — cluster machine.id and label key/values were interpolated unquoted into the remote SSH command; added charset validation.

Rejected findings (verified, deliberately not filed)

• dora logs raw .txt files bypass all filters — real (Medium), but the correct fix synthesizes timestamps/levels and changes output ordering; a behavior change that needs maintainer design input, not an autonomous patch.
• dora cluster upgrade reconnect check false-positive — a stale still-connected daemon can satisfy the post-restart wait; real, but a correct fix (disconnect-then-reconnect / connection-identity tracking) is design-level and too speculative to patch blind.
• dora cluster up coordinator addr 0.0.0.0/loopback unvalidated for multi-machine clusters — plausible guard but borderline feature-add; low confidence it isn't intentional.
• --hub-override split_once('=') mis-parses an =1.2.3 version in the key — low; only yields a confusing error, and a version in the override key is meaningless anyway.
• Duplicate --hub-override keys silently overwrite — very low footgun.
• dora top stale table selection not re-clamped after the node list shrinks — cosmetic; ratatui clamps at render time, no panic.
• dora hub publish version auto-detect fails for version.workspace = true crates — fails gracefully with an actionable "pass --version" message; a real fix requires resolving the workspace-root Cargo.toml at the committed rev.
• dora hub fetch dataflow-vs-package detection is extension-case-sensitive (.YAML) — cosmetic UX only.
• merge_env doc-vs-behavior mismatch; ROS2 multi-topic validation gap (core) — not correctness bugs (cosmetic / caught later by the bridge binary).
• mavlink2-bridge / arrow-convert / node & operator APIs — reviewed field-by-field; no solid findings. This code is heavily hardened with regression history.

Validation

Per-PR: cargo fmt --all -- --check, cargo clippy -p dora-cli --all-targets -- -D warnings, and cargo test -p dora-cli all green. Combined diff additionally run through the /simplify gate (which prompted collapsing the #2859 helper into the canonical Display and broadening #2861 from id-only to labels). The full workspace test suite is green except for 4 pre-existing zenoh/ros2 networking tests that require IPv6 (tcp/[::]:0 — unavailable in this sandbox) and are unrelated to these CLI-only changes.

---

Generated by Claude Code

[phil-opp]

Automated review by Claude — this is a fully automated review; no human has vetted it.

No issues found. Delegating to InputMapping's Display is the correct fix: it picks the coarsest exact unit and produces the canonical source string the descriptor parser round-trips, so dora/timer/hz/2000 (500µs) no longer collapses to dora/timer/millis/0. The removed User/Logs match arms rendered identically to Display, so this is behavior-preserving for those cases while removing a small reimplementation.

---

Generated by Claude Code

[github-actions]

@phil-opp the Trunk merge queue failed for this PR.

See the Trunk merge-status comment for details.

Posted as a new comment so GitHub sends an email — Trunk's sticky comment is edited in place and won't trigger a notification.