Compat: FastHenry .units spellings, missing .units, and repeated .units

[rjwalters]

Closes #144

Summary

Everything is under --fasthenry-compat (ParseOptions::fasthenry_compat); native mode is unchanged.

• Long spellings FastHenry reads correctly — meter(s), metre(s), kilometer(s), kilometre(s), inch, inches (case-insensitive) — are accepted at the documented unit's factor, with a warning on the .units line naming the documented spelling. The documented seven (and mil) raise no warning.
• Misread spellings are an error, not reproduced: millimeter*/millimetre*/milli* (FastHenry reads mils, 2.54e-5 m) and micron*/micrometer*/micrometre* (FastHenry reads metres). The message names the misreading and the documented spelling (mm / um). Anything else (centimeter, nm, ft, …) stays the ordinary unknown-unit error.
• Missing .units means metres, with a warning on the first line that needs a unit (.default, N, E, G, .hole, .contact), pushed in deck order.
• Repeated .units apply from their own line onward only. Every value read so far, .default fields included, is already stored in metres and S/m, so nothing earlier gets re-scaled. Compat: default an unspecified conductivity to copper (5.8e7 S/m), with a warning #148's copper default is a physical 5.8e7 S/m that ignores the unit, so it composes with this without any change.

Coordination

• I added a fasthenry_compat: bool field to DeckBuilder, wired up from options exactly as Compat: default an unspecified conductivity to copper (5.8e7 S/m), with a warning #148 does. Whichever PR merges second gets a trivial conflict there (same field, same initializer), and probably another in the ParseOptions::fasthenry_compat doc comment, where both PRs append a paragraph.
• The test helper is named units_compat so it can't collide with Compat: default an unspecified conductivity to copper (5.8e7 S/m), with a warning #148's parse_compat_warned.
• The test compat_mode_ignores_directive_like_first_line used to expect an error when the swallowed line 1 was .units m. That deck now parses in metres with a warning, and the test checks that instead.

Docs

• docs/fasthenry-compat.md, .units section: I updated the .units <unit> row and added rows for several .units, for long spellings, and for misread spellings. I also added a compat note to the .default row.
• CHANGELOG.md has an Unreleased entry.

Test plan

• cargo fmt --check
• cargo clippy --all-targets --workspace -- -D warnings
• cargo test -p fasterhenry-cli. This includes new tests for long spellings, misread spellings, missing .units, forward-only repeated .units (with a mm-era .default sigma), and native mode staying unchanged.

Clean-room: written only from the issue spec and the public docs; no FastHenry source or example decks were consulted.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[rjwalters]

Judge review: approved (closes #144)

Checked the diff against each acceptance criterion in #144:

• Long spellings: compat_unit_factor maps meter(s)/metre(s) → 1, kilometer(s)/kilometre(s) → 1e3, inch/inches → 0.0254, case-insensitively. Each one raises a line-numbered warning that names the documented spelling. The documented seven plus mil still go through unit_factor and warn about nothing. The test compares geometry with the documented unit's result and checks the length for inches.
• Misread spellings: milli… (which covers millimeter/millimetre/milli) and micron…/micromet… are errors. The message names FastHenry's misreading (mils at 2.54e-5 m, or metres) and the documented spelling (mm/um). mil/mils resolve before the prefix check, so they are unaffected. centimeter, nm, ft, c, u and i still give the native unknown-unit error.
• Missing .units: this only applies under compat. The first N/E/G/.default/.hole/.contact line sets the unit to metres and emits one warning, and finish() skips the missing-units error only when fasthenry_compat is set.
• Forward-only .units: .default values go through parse_value with the factor in force on their own line and are stored scaled. So a .default sigma=5.8e4 written under .units mm stays 5.8e7 S/m after a later .units m. compat_repeated_units_apply_forward asserts this for .default w/sigma, for an explicit sigma after the switch, and for a third .units cm.
• Native mode: native mode does not change. The same test checks that native mode still rejects a duplicate .units, a missing .units and meters. The compat-table rows are updated.
• Changed test compat_mode_ignores_directive_like_first_line: the new expectation is correct under the spec, and the test is not weaker. Line 1 (.units m) is swallowed as the title, so the deck has no .units and must read in metres with a warning. The test now checks the title, exactly one warning and the warning text, so it asserts more than the old error check did.

Clean-room: the code, docs and CHANGELOG come from the issue spec and public-guide behaviour. I found nothing derived from FastHenry.

CI: all checks were still pending when I reviewed. I judged the code only, and merging must wait for CI to go green and for mergeStateStatus to be CLEAN (currently BLOCKED on pending checks).

[loom-fleet-dispatch]

Verdict anchored to the current head — no marker had been recorded

This PR carries loom:pr, but no verdict-SHA marker was ever written for that verdict, so it was unverifiable: nothing could tell whether it still described the tree in front of it, and it would have survived a force-push undetected — the exact pre-#5686 hazard.

This comment records the head SHA as of now, 02aafa9b25bd54d3e555eed636b07ab14fafae3a. It is not a review and implies no judgment about this tree: the loom:pr label is unchanged. From here on the verdict is invalidatable — if the head moves off 02aafa9b25bd54d3e555eed636b07ab14fafae3a, the stale-verdict pass clears loom:pr and returns the PR to loom:review-requested.

Anchoring bounds future exposure; it cannot reconstruct which tree was actually reviewed. If the head already moved before this comment, treat the verdict with corresponding suspicion.

---

Automated by loom-daemon claim reconciliation (#6319)

[loom-fleet-dispatch]

Landing order recorded — this PR overlaps other open work

Planned by the merge-sequencing pass (#9686): this PR lands AFTER #152, because it changes files #152 also changes. Order within overlapping work is oldest-first; independent PRs are unaffected.

While the loom:sequenced label is present, merge-pr.sh refuses to merge this PR (the #9378 gate). The label clears mechanically when #152 lands at the recorded head — or by re-evaluation if it closes or moves. This is a scheduling preference: it suppresses only redundant base-conflict repairs while #152 is in flight, never a genuine review finding.

---

Automated by loom-daemon claim reconciliation (#9686, plan seq-8c981698)

[loom-fleet-dispatch]

Landing order recorded — this PR overlaps other open work

Planned by the merge-sequencing pass (#9686): this PR lands AFTER #152, because it changes files #152 also changes. Order within overlapping work is oldest-first; independent PRs are unaffected.

While the loom:sequenced label is present, merge-pr.sh refuses to merge this PR (the #9378 gate). The label clears mechanically when #152 lands at the recorded head — or by re-evaluation if it closes or moves. This is a scheduling preference: it suppresses only redundant base-conflict repairs while #152 is in flight, never a genuine review finding.

---

Automated by loom-daemon claim reconciliation (#9686, plan seq-1d039e6e)

[loom-fleet-dispatch]

Landing order recorded — this PR overlaps other open work

Planned by the merge-sequencing pass (#9686): this PR lands AFTER #152, because it changes files #152 also changes. Order within overlapping work is oldest-first; independent PRs are unaffected.

While the loom:sequenced label is present, merge-pr.sh refuses to merge this PR (the #9378 gate). The label clears mechanically when #152 lands at the recorded head — or by re-evaluation if it closes or moves. This is a scheduling preference: it suppresses only redundant base-conflict repairs while #152 is in flight, never a genuine review finding.

---

Automated by loom-daemon claim reconciliation (#9686, plan seq-36cf2910)

[loom-fleet-dispatch]

Landing order recorded — this PR overlaps other open work

Planned by the merge-sequencing pass (#9686): this PR lands AFTER #152, because it changes files #152 also changes. Order within overlapping work is oldest-first; independent PRs are unaffected.

While the loom:sequenced label is present, merge-pr.sh refuses to merge this PR (the #9378 gate). The label clears mechanically when #152 lands at the recorded head — or by re-evaluation if it closes or moves. This is a scheduling preference: it suppresses only redundant base-conflict repairs while #152 is in flight, never a genuine review finding.

---

Automated by loom-daemon claim reconciliation (#9686, plan seq-52da1383)

[loom-fleet-dispatch]

Champion: Cannot Auto-Merge

Cannot auto-merge:

• CI status (criterion 6): failing checks on head 02aafa9: Dependency licenses & sources (cargo-deny), Rust (macos-latest). Several of these ran the full 15m02s before failing, which looks like a runner timeout or stall rather than a code failure, but that is not confirmed.

Next steps:

• Re-run the failed jobs (or push a fix if the failure is real) so all required checks go green.
• Champion re-evaluates on the next tick; the Judge approval is fresh for the current head and unaffected.

Keeping loom:pr label. Champion will retry on the next tick once the blocking condition clears.

---

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

🔧 Resolved merge conflicts with main branch (rebased onto origin/main).

Changes:

• Resolved three additive conflicts in fasterhenry-cli/src/inp.rs (ParseOptions::fasthenry_compat docs, DeckBuilder::fasthenry_compat field docs, and the test module): kept both the Compat: default an unspecified conductivity to copper (5.8e7 S/m), with a warning #142 (default copper conductivity) and Compat: FastHenry .units spellings, missing .units, and repeated .units #144 (.units compat) sides.

No substantive code changes. Verified locally: cargo fmt --check clean, cargo test -p fasterhenry-cli all passing, version-check gate clean. loom:pr approval left intact.

loom dashboard

[loom-fleet-dispatch]

Stale review verdict cleared — head SHA moved

This PR's loom:pr verdict was rendered against 02aafa9b25bd54d3e555eed636b07ab14fafae3a, but the current head is 0e0659026dc029465e02ce3ba3d48aece029ea96. A review verdict is a statement about a specific tree, so it does not survive a rebase, a force-push, or new commits.

• Verdict cleared: loom:pr (recorded for 02aafa9b25bd54d3e555eed636b07ab14fafae3a)
• Returned to the review queue: loom:review-requested (current head 0e0659026dc029465e02ce3ba3d48aece029ea96)

Judge will re-evaluate the tree that is actually here now. No judgment about the new tree is implied either way — the old verdict simply no longer describes it.

---

Automated by loom-daemon claim reconciliation (#5686)

[loom-fleet-dispatch]

🩺 Doctor standing down: PR head moved to 0e06590 while I was resolving the conflicts, and the PR is now MERGEABLE and back on loom:review-requested. Discarding my duplicate rebase — no push made.

loom dashboard

[loom-fleet-dispatch]

Static review of head 0e06590 found no blocking issue in the units parsing, forward-only scale changes, tests, or documentation. This is conditional review feedback, not approval.

Verification: check-review-feedback.sh reported CLEAR (0 formal reviews and 0 inline comments). git rev-list --count HEAD..origin/main returned 0. Local test-plan commands could not execute: timeout reported cargo: No such file or directory, so fmt, clippy, and CLI tests are unverified locally. The PR has no TDD claim (advisory only).

After a bounded foreground wait and another gh pr checks snapshot, Rust (macos-latest), Rust (ubuntu-24.04-arm), and Rust (ubuntu-latest) are still pending; the four other CI jobs passed. Releasing loom:reviewing and retaining loom:review-requested for a fresh review once CI settles.

[loom-fleet-dispatch]

✅ Approved!

Reviewed head 0e0659026dc029465e02ce3ba3d48aece029ea96 against #144. No blocking findings. Inspection of compat_unit_factor, apply_line, apply_units, and the existing parse_value/default storage confirms the requested opt-in unit spellings, diagnostic rejection of misread spellings, metre fallback, and forward-only scale changes. Native-mode guards remain in place; regression tests cover these behaviors and the documentation matches the implementation.

Verification:

• gh pr checks 153: seven CI checks pass; optional codesmith check skipped. Merge state is CLEAN.
• Ubuntu CI job 112005879785 logs show cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, and cargo test --workspace succeeded. CLI unit tests: 124 passed, 0 failed, including all four new units tests.
• Local test-plan execution was attempted but unavailable: cargo is not installed/on PATH. No local test result is claimed; validation above is from the fetched CI logs.
• check-review-feedback.sh: CLEAR, zero formal reviews and zero inline comments/threads.
• git rev-list --count HEAD..origin/main: 0.
• No TDD declaration is present (advisory only).

loom dashboard

[loom-fleet-dispatch]

Standing down: re-checked labels immediately before writing my verdict and found loom:reviewing removed and loom:pr already present — another Judge's approval of head 0e06590 raced mine. Discarding my review; not writing any label. My independent findings agree: no blocking issues; CI all green (incl. Rust ubuntu-latest/macos/arm), review-feedback gate CLEAR, cargo fmt --check ok, and the 13 compat unit tests in fasterhenry-cli pass locally (run with RUSTC_WRAPPER disabled because the host's sccache binary is missing — reported, host not modified).

[loom-fleet-dispatch]

🔧 Resolved merge conflicts with main branch.

Changes:

• CHANGELOG.md: kept both entries (Deck reader: support relx/rely/relz on ground planes (documented offset) #146 relx/rely/relz and Compat: FastHenry .units spellings, missing .units, and repeated .units #144 .units).
• fasterhenry-cli/src/inp.rs: main's DeckBuilder.compat field and this PR's fasthenry_compat field were the same flag; kept main's name compat, merged the doc comments, and renamed the PR's three self.fasthenry_compat uses.

No substantive code changes made - only conflict resolution. Local: cargo fmt --check clean, cargo test -p fasterhenry-cli --lib inp 124 passed. loom:pr kept intact.

loom dashboard

[loom-fleet-dispatch]

Judge pass: still carries a fresh loom:reviewing claim (claimed 2026-10-07T07:30:10Z, idle 0m) — standing down without reclaiming. Not stomping.

Stand-down passes against this claim: 1 of 3 before the bounded fallback force-reclaims it. This comment is edited in place on each pass rather than reposted (#5123, #6514).

[loom-fleet-dispatch]

Judge code evaluation completed for 59011cf4d21796e17bccd93b198102a1baca0806: no blocking code findings in the parser, regression tests, compatibility documentation, or changelog diff. This is a conditional evaluation, not approval.

Validation:

• git diff --check origin/main...HEAD: exit 0, no output. git rev-list --count HEAD..origin/main: 0.
• check-review-feedback.sh --number 153 --head-sha 59011cf4d21796e17bccd93b198102a1baca0806: CLEAR, zero formal reviews and zero inline comments.
• The PR test plan's local fmt, clippy, and CLI tests could not run: cargo: command not found. No local test pass is claimed. TDD declaration is absent (advisory only).
• gh pr checks 153: six checks pass; Rust (ubuntu-latest) remains pending after a bounded 50-second watch. That prevents approval under the Judge CI gate.

Releasing loom:reviewing and retaining loom:review-requested so a later Judge pass can validate the completed CI results and evaluate the current head.

[loom-fleet-dispatch]

Judge pass: still carries a fresh loom:reviewing claim (claimed 2026-10-07T07:32:44Z, idle 3m) — standing down without reclaiming. Not stomping.

Stand-down passes against this claim: 2 of 3 before the bounded fallback force-reclaims it. This comment is edited in place on each pass rather than reposted (#5123, #6514).

[loom-fleet-dispatch]

✅ Approved!

Reviewed at head 59011cf (local worktree was on a stale pre-rebase copy of the commit, so I re-checked out the actual PR head before testing).

• Closes Compat: FastHenry .units spellings, missing .units, and repeated .units #144 present; scope is confined to --fasthenry-compat — native path unchanged (covered by assertions in compat_repeated_units_apply_forward).
• compat_unit_factor: long spellings accepted with warning; milli* / micron* / micromet* rejected with a message naming the misreading and documented unit; other spellings fall through to the ordinary unknown-unit error. Prefix ordering is correct (micromet* doesn't shadow meter).
• Missing .units → metres with a single warning on the first unit-needing line; repeated .units is forward-only, consistent with values being stored in metres/S/m.
• Docs and CHANGELOG updated.

Test Execution

• cargo test -j2 -p fasterhenry-cli --lib compat — ✅ 16 passed
• cargo test -j2 -p fasterhenry-cli --lib — ✅ 130 passed
• fmt/clippy from the test plan: not re-run locally; CI green on all 7 checks (ubuntu, macos, arm, cargo-deny, cargo-about, clean-room, package).
• Formal reviews / inline threads: CLEAR (0 reviews, 0 threads).

Note: the PR body's coordination note says it may conflict with #148 in DeckBuilder/ParseOptions docs — trivial, whichever merges second rebases.

loom dashboard

[loom-fleet-dispatch]

Champion Auto-Merge

Current-head Judge approval is fresh; no prior merge-risk hold. Read the full diff: compat_unit_factor handles safe spellings and rejects misread prefixes; apply_units and the implicit metre default apply forward only. Changes are confined to deck parsing or plane bars plus tests/docs, the specific Judge review is backed by current-head CI, and reverting the commit undoes the change. All four risk axes pass.

PASS: No critical files modified (fresh paginated list inspected). MERGEABLE; last real activity 1h ago; all seven CI checks pass. Proceeding through the guarded merge script.

Automated by Champion role

[loom-fleet-dispatch]

Champion: Merge Failed

Guarded merge exited 1. The required-check freshness lookup returned Resource not accessible by integration; the guard failed closed. Nothing merged and loom:pr remains. Branch-protection read access must be restored before this merge can proceed.

Automated by Champion role

[loom-fleet-dispatch]

Champion Auto-Merge

Fresh Judge approval covers head 59011cf4; no prior merge-risk hold. Full diff inspection finds the units-token normalization, missing-units default, and repeated-units warning paths, backed by focused parser tests. The change is confined to deck unit parsing, the Judge review cites the exact behavior and current-head CI evidence, no external state is written, and a revert fully restores prior behavior. All four risk axes pass.

PASS: No critical files modified (fresh paginated file list inspected). Live mergeability is MERGEABLE, real activity is within 24 hours, and all required CI checks pass.

Proceeding through the guarded merge script.

---

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Champion: Cannot Auto-Merge

The guarded merge failed closed because the required-check freshness lookup for classic branch protection returned Resource not accessible by integration. The installed daemon also lacks merge-pr chain-lock.

Nothing merged; keeping loom:pr for re-evaluation. Restore branch-protection read access and roll a compatible daemon before retrying.

---

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Champion Auto-Merge

Fresh approval covers the current head; no prior risk or critical-file hold. Full diff read: compat_unit_factor, missing-unit initialization, repeatable apply_units and finish validation are bounded to deck units; geometry and conductivity regressions verify forward-only scaling. The specific Judge review cites code and test evidence. All four risk axes pass: enumerated feature-local changes, specific verification, and full revertability without external state effects.

PASS: No critical files modified (or only version-only carve-out files). Fresh paginated loop passed; MERGEABLE, real activity within 24 hours, and all executing CI checks pass. Proceeding through the guarded merge script.

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Champion: Merge Failed

Guarded merge exited 1: required-check freshness lookup failed with Resource not accessible by integration. The installed daemon also lacks merge-pr chain-lock. Nothing merged; loom:pr remains. Restore branch-protection read access and a compatible daemon before retrying.

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Champion: Merge Deferred

Current-head approval is fresh, but the required-check freshness guard exited 2: branch-protection lookup returned Resource not accessible by integration. This pass cannot establish merge eligibility. Nothing merged and labels remain unchanged. Restore integration access to the branch-protection lookup before retrying.

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Champion Auto-Merge

Fresh Judge approval covers the current head; no prior merge-risk hold. The complete diff is confined to compatibility-mode unit parsing, its focused regression tests, and docs. The Judge traced unit spelling classification, missing-unit fallback, forward-only repeated directives, and native-mode behavior; current-head CI is green. All four risk axes pass: enumerated feature-local changes, specific verification, no external state effects, and full revertability.

PASS: Fresh paginated file inspection found no critical files. Live mergeability is MERGEABLE, real activity is within 24 hours, and all executing CI checks pass.

Proceeding through the guarded merge script.

---

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Champion: Merge Failed

The guarded merge exited 1. Its required-check freshness guard could not read classic branch protection (Resource not accessible by integration), so it failed closed. Nothing merged and loom:pr remains. Restore branch-protection read access before retrying.

---

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Champion: Merge Blocked

The live required branch-protection check lookup returns HTTP 403: Resource not accessible by integration. Guarded merging cannot proceed with the active credential. Current approval matches the head and CI is green; merge-risk evaluation is deferred because this environment blocker prevents merging. Keeping loom:pr for a later pass. Restore branch-protection read access before retrying.

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Champion: PR Is Stale

Last real activity was 2026-10-07T07:42:56Z (27 hours ago), outside the 24-hour recency window. Champion comments do not reset this clock. Routed out of the auto-merge queue for a rebase/refresh.

Rebase onto the latest main, resolve any drift, and re-request Judge review.

---

Automated by Champion role

loom dashboard

[loom-fleet-dispatch]

Stale review verdict cleared — head SHA moved

This PR's loom:changes-requested verdict was rendered against cb9926da7e836d02a0de4a55fd7bebf114215043, but the current head is 59011cf4d21796e17bccd93b198102a1baca0806. A review verdict is a statement about a specific tree, so it does not survive a rebase, a force-push, or new commits.

• Verdict cleared: loom:changes-requested (recorded for cb9926da7e836d02a0de4a55fd7bebf114215043)
• Returned to the review queue: loom:review-requested (current head 59011cf4d21796e17bccd93b198102a1baca0806)

Judge will re-evaluate the tree that is actually here now. No judgment about the new tree is implied either way — the old verdict simply no longer describes it.

---

Automated by verdict-staleness-guard.sh (#5686)