chore: make the binary-size check measure the pull request's own code

[jdx]

https://entire.io/gh/jdx/usage/trails/50

The size check has not been measuring pull requests. Every recent report shows exactly +0 bytes on every binary, including #1497, which adds a lot of code to usage-lib and the CLI. #1498 then failed the check with a compile error that its own build doesn't have (no field delegate on type SpecComplete).

Cause. tasks/size.sh builds the merge base (in a worktree) and the head into one target directory, so crates.io dependencies compile once. But Cargo names a workspace crate's artifacts by its path relative to the workspace root, so both checkouts produce the same artifact names. After the base build, the head build found a fresh usage-lib under the same name and reused it. The head binary was therefore built from the base's library code. That explains the +0s, and the #1498 failure: its usage-cli rebuilt (it adds a libc dependency) against the base's usage-lib, which has no delegate field. The same reuse can happen across CI runs through the rust-cache'd target directory, and locally across branches.

Fix. Before each build, the script now runs cargo clean --release -p <member> for every workspace member, taken from cargo metadata --no-deps. Each side's workspace crates always build from that side's source, and crates.io dependencies stay shared and cached. The 1% gate, the binaries measured and what is gated are unchanged.

Validation. Run locally against the two open pull requests, reusing one target directory across both runs on purpose, since reuse is what broke:

PR	usage CLI	derived mise CLI (usage's share)
#1497	+9,272 B (+0.08%)	+0
#1498	+79,424 B (+0.69%) at f4d08fea91; +31,808 B (+0.28%) after 1e065f682c dropped an mpsc channel	+0

Before this change, the #1497 run in that setup failed to compile the same way #1498 did in CI. Both pull requests are under the gate. The derived CLI doesn't move because neither touches the derive or argv runtime.

🤖 Generated with Claude Code

---

Note

Low Risk
Touches only the tasks/size.sh CI script; no runtime, auth, or application code paths.

Overview
Fixes the binary-size CI check so head and base builds are compared against their own workspace crate artifacts instead of silently reusing each other’s from a shared CARGO_TARGET_DIR.

In tasks/size.sh, build() now discovers workspace members with cargo metadata --no-deps and runs cargo clean --release for each member before cargo build, while still sharing the target directory for crates.io deps. Comments are updated to document why the clean is required (identical artifact paths across the base worktree and the PR checkout caused +0 byte deltas and wrong linkages, e.g. head CLI against base usage-lib).

The 1% gate, measured binaries, and gating rules are unchanged—only measurement correctness.

^{Reviewed by Cursor Bugbot for commit aba852a. Bugbot is set up for automated code reviews on this repo. Configure here.}

Summary by CodeRabbit

• Chores
  • Build size comparisons now use artifacts from the correct checkout, avoiding misleading results caused by reusing cached build outputs. This makes reported size differences more accurately reflect the code being compared.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Central YAML (base), Organization UI (inherited)

Review profile: CHILL

Plan: Advanced

Run ID: f1223b24-c502-4d5c-846d-92334fe12b9d

📥 Commits

Reviewing files that changed from the base of the PR and between 373bfb9 and aba852a.

📒 Files selected for processing (1)

• tasks/size.sh

Included review availability: Your plan provides up to 4 included reviews per hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

The size comparison build now identifies workspace packages with cargo metadata and cleans their release artifacts before building. The updated comment explains that separate checkouts can otherwise reuse artifacts and report a comparison of +0 bytes.

Changes

Size comparison

Layer / File(s)	Summary
Clean workspace release artifacts tasks/size.sh	The build step gets workspace package names from cargo metadata --locked --no-deps and runs cargo clean --release --locked for those packages before the release build. The comment describes the artifact reuse issue.

Estimated code review effort: 2 (Simple) | ~8 minutes

Change: Bug fix · Severity of issue fixed: Low

Merge Risk: ⚪ Minimal · up to aba85

The size comparison cleans workspace release artifacts before each checkout’s build, helping prevent stale artifacts from skewing results. No concrete merge-blocking risk is established; the PR is ready for normal checks.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 1 functions across 1 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: ensuring the binary-size check measures the pull request's own code.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

• Fix all pre-merge checks with AI

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge; no outstanding finding or new actionable issue remains.

Summary

The PR makes the binary-size check rebuild workspace crates from each checkout while retaining a shared target directory. The follow-up change applies --locked to metadata and clean, matching the build’s lockfile constraint.

_{Reviews (2) · Last reviewed commit: "chore: keep the size check from rewritin..."}

[github-actions]

Instruction counts

benchmark	trend	instructions	Δ	wall (min)	Δ
markdown	▆▇▆▇▇█▁▂	285,027,329 → 285,064,854	+0.01%	49.92 → 49.49ms	-0.86%
startup	▁▃▃▃▆▆▂█	1,053,355 → 1,060,174	+0.65%	1.46 → 1.47ms	+0.84%

No instruction-count regression above 1%.

_{Only instruction counts gate. Wall clock is shown for context — on identical hardware it moves 4-20% run to run.}

_{Measured by tak — instruction-counted CLI benchmarks, stored in this repository's git notes.}

Shadow comparison

Parsing mise use -g node@20 against a shadow of mise's committed spec.
Reported, not gated: the shadow grows as the derive learns to express more, so
what to watch is the ratio rather than either column.

framework	stripped binary, bytes
usage	1262000
bpaf	2494656
clap	3100904

framework	instructions, cold parse	vs usage
usage	8530	—
clap	6314978	740x
bpaf	21909298	2568x

                                              min       p01       p10    median
usage-rs: argv -> struct                      793       797       801       831  ns
clap: build tree + parse -> struct        1243479   1246841   1254212   1271254  ns
bpaf: build parser + parse -> struct      3415140   3415140   3478484   3500207  ns

usage: argv -> struct                             763 ns      0.76 µs
clap: build tree + parse -> struct            1259890 ns   1259.89 µs
clap: parse -> struct, tree reused              50897 ns     50.90 µs
clap: build tree only                          749413 ns    749.41 µs

_{aba852a706ac vs 373bfb909efd · measured on the runner, not pushed to the history.}