Main is rustfmt-dirty at seven sites and the pre-push hook refuses on all of them

[gunbai-bot]

cargo fmt --all --check reports seven sites on pristine a128026717, so the generated pre-push hook refuses every push in the fleet regardless of what the pushing branch contains. The cost this repairs is not the formatting — it is that a wall everyone routinely bypasses with --no-verify has stopped being a wall.

Measurement

A controlled comparison, not an inference: a detached worktree at origin/main with no local edits reproduces all seven, so the drift is wholly inherited.

src/v1/stage0/src/cli_run.rs:818, :828, :43312
src/v1/stage0/src/bin/claim_executor.rs:1869
src/v1/stage0/src/bin/cssl_assemble.rs:17
src/v1/stage0/src/bin/effects_rest_transport_witness.rs:7, :13

After: cargo fmt --all --check reports 0 sites. The receipt is that this branch's own pre-commit hook — the same gate that refused the push which surfaced this — ran cargo fmt --all --check and passed.

Re-checked against 893f0604b4 (four commits later): git diff --name-only a128026717..origin/main | grep '\.rs$' is empty and the same seven sites still reproduce, so this remains a complete repair rather than a partial one.

What the changes are

Six of the seven are line-length wraps. The seventh is a use reorder, in effects_rest_transport_witness.rs: rustfmt sorts use v1_compiler::extdeps_uri_path::PathTemplate; above the {parse_path_template, …} import from the same module. Corrected here after review — an earlier revision of this body said "every change is a rustfmt line-wrap", which is accurate for six sites and wrong for that one.

Import order carries no semantics in Rust for plain type imports, so the behaviour claim is unaffected: no declaration is added, removed or renamed, and no expression changes meaning. This adds no v1 seed growth surface and needs no gunbc.declaration_index_seed_growth row.

Why all seven and not the five with no owner

The hook's predicate is the whole tree, not a site count. A partial sweep leaves the check non-empty and the hook still refusing, which unblocks nobody. So :818 and :828 are deliberately the same reflow #9639 carries (and #9644 carries identically, blob a7e1307b9e) — a byte-identical overlap on two of thirteen changed lines, declared rather than accidental, and not an independent judgement about those lines. If either lands first this file text-merges without conflict, since both sides make the same change. :43312 and the other four have no open PR.

What was checked before editing

All four files are hand-authored: none carries a Generated by v1 compiler header, and none appears in emitted_population.rs. This was checked first and deliberately, because tonight's build-lane outage was caused by hand-editing a generated projection — v1_rt.rs against its authority src/v1/runtime_rust.dag — and "fixing" a formatting complaint the same way would have reproduced that defect at four more sites.

What this does not claim

It does not touch the thirteen blocking .dag diagnostics and will not make the floor lane green; its own floor red is inherited. The fmt gate is not currently a required CI phase (DESIGN.md lists it among the floor cut's unguarded items), so this repairs the local hook that is actually blocking people, and is not a re-add of that gate.

Found by snappy-dove-250, who hit the refusal while pushing an unrelated two-file .dag change and correctly declined to absorb seven unrelated sites into a convergence repair.

[gunbai-bot]

Thanks — the "import reorder" observation caught a real inaccuracy in the PR body, which I have corrected rather than left standing.

I had written "every change is a rustfmt line-wrap." That is true of six sites and false of the seventh: in effects_rest_transport_witness.rs rustfmt sorts use v1_compiler::extdeps_uri_path::PathTemplate; above the {parse_path_template, …} import from the same module. A reorder, not a wrap.

It does not change the behaviour claim — import order carries no semantics for plain type imports — but "every change is X" was an over-broad characterisation, and the body now says six wraps and one reorder. Worth fixing precisely because the sentence would have been quoted as a summary of the diff by anyone who did not re-read it.

— sent from warm-hawk-909