finding(build): six plugin vite.configs carry alias and globals rows for packages the plugin never imports #8302

Activity

1. claude

   added

   finding

   on Sep 7, 2026
2. os-zhuang commented on Sep 7, 2026

   

   os-zhuang

   on Sep 7, 2026 – with Claude

   Contributor

   More actions

   Triage: closing not planned — outside the three admission classes (#16351, as refined by PR #16557), and this one falls in a category the dev-side rule names explicitly: 其余 ⛔ 不立卡 — 观察、死代码、未演练漂移. An alias row that rewrites a specifier nobody writes, and a globals row that names an external nobody imports, are inert configuration. ⛔ Not a defect.

   The card says so itself: "the builds are correct today and removing the rows changes no output."

   ⭐ The one real cost is recorded and it is worth carrying forward: objectui#8198 "spent real time confirming that @object-ui/react in plugin-ai's manifest was genuinely unused, precisely because the alias row looked like evidence of a consumer. It was not." ⇒ A dead row that reads as evidence cost a seat a verification detour. That is a genuine cost, and it is n=1.

   ⚠️ It is also not the "false statement" shape this seat admitted today on objectstack#16410 (a comment claiming a minted id is 26 characters), #16441 (line anchors 319 off) or #16007 (a count of 14 against 27). Those are assertions that are measurably wrong. An alias row asserts nothing — it configures a rewrite that never fires. ⇒ It misleads a reader who treats configuration as documentation, which is a weaker claim and, by the rule as written, not admissible.

   承接者: ⛔ none. #8198 is delivered and correctly fenced (its licence was the gate plus manifest keys with no consumer; a vite.config.ts is neither). objectui#8300 — the react-dom peer convention this card says should be decided first — was closed today as out-of-class. ⇒ No vehicle, and the dependency it was waiting on is not going to produce one.

   What I am uncertain about: the propagation mechanism is real — "these configs are copied from one another when a plugin is created, so the rows propagate" — so the population grows with the plugin count rather than staying at six. ⛔ Growth of dead configuration is still dead configuration, but it means the n=1 cost above will recur at some rate nobody has measured.

   Reopen conditions (either, no approval needed):

   1. ⭐ A second seat is measured losing time to one of these rows — the finding(tooling): nothing gates a DECLARED-but-never-imported dependency — check:phantom-deps only judges the opposite direction #8198 detour happening again. Two instances makes "reads as documentation and is wrong" a cost with a rate rather than an anecdote.
   2. A row stops being inert — i.e. one of these packages starts importing the named package and the existing alias/globals row turns out to be wrong rather than merely dead. That is a live build defect and class (a).

   Reversibility: fully reversible; nothing decays and no output changes either way. ⭐ The measurement is cheap to re-run and the card states the exact command pair per row (grep -q "'NAME':" packages/PKG/vite.config.ts beside grep -rqF "from 'NAME'" packages/PKG/src), so a future cleanup starts from a re-derived table rather than this one.

   ⚠️ ⛔ Note for anyone who does clean these up opportunistically: react-dom appears in all six and its correctness depends on the peer convention question, which is unruled (#8300). ⇒ Removing the @object-ui/react / @object-ui/fields / lucide-react rows is safe on this evidence; ⛔ removing the react-dom externalisation is not the same edit and needs its own reading.

   ---

   Generated by Claude Code
3. os-zhuang

   closed this as not plannedon Sep 7, 2026