The lint stage went green earlier this week, but the script still failed
locally on the cargo-machete step it runs afterwards — so anyone running
it saw red and learned to ignore it, which is the same problem the lint
work existed to fix.
Every removal was verified before it was made. cargo-machete is a text
search and will flag a dependency reached through a macro, a re-export,
or a cfg-gated path, so each of the 19 was grepped for `use X` and `X::`
in its own crate first; all came back with nothing.
Nine were ours. `agent_joule_panel` still declared jcp-bridge,
joule-code-jcp, joule-code-joulecontract, joule-code-release,
joule-code-spec, jouleclaw-runtime, joulecontract-schema, ignore and lsp
— fork-evolution leftovers, each one a crate the panel paid to build and
never touched.
The `visual-tests` feature goes too, which I had left alone when its
binary was removed because I could not then rule out breaking Linux.
Reading every `image` declaration settled it: image is a *regular*
dependency on Linux and FreeBSD through a target section, and `dep:image`
is also pulled in by the live `test-support` feature. So `visual-tests`
was the only consumer of tempfile, action_log and clock/test-support, and
dropping it orphans nothing.
The `collab` failure was a different thing wearing the same hat: not an
unused dependency but a manifest that would not load. `crates/collab` is
not a workspace member and nothing depends on it, yet its own manifest
declares `collab = { workspace = true }` against an entry this fork
deleted from [workspace.dependencies] — so machete died loading a
manifest nothing builds. The entry is restored, as upstream has it;
membership is unchanged.
Which leaves `crates/collab` itself: 95 tracked files, outside the
workspace, referenced by nothing. The same shape as the visual test
runner, and left in place rather than removed on my own judgement.
Co-Authored-By: Claude Opus 5 <[email protected]>
94e626045f