João Salavisa
97 lines
4.6 KiB
JavaScript
97 lines
4.6 KiB
JavaScript
/**
|
|
* Shape checks on the i18n guard allowlist itself.
|
|
*
|
|
* Split out of `check-guard-allowlist.mjs` so they can be tested directly: that
|
|
* script resolves this repository's own git history and options module from
|
|
* fixed paths, so it cannot be pointed at a fixture.
|
|
*/
|
|
|
|
/**
|
|
* Entries listed more than once, each named once however often it repeats.
|
|
*
|
|
* Nothing else looks for one, because a duplicate changes no behaviour: ESLint is
|
|
* indifferent to a repeated pattern, and the removal check in
|
|
* `check-guard-allowlist.mjs` only ever inspects entries that LEFT the array. So
|
|
* when #2214 added two, lint, `i18n:check`, the ratchet and the unit suite all
|
|
* passed and it was caught by eye. What a duplicate damages is the record: a
|
|
* second copy of a path an earlier migration already listed reads as the new
|
|
* PR's own coverage, which is the misinformation the comments around the list
|
|
* exist to prevent.
|
|
*
|
|
* EXACT duplicates only, deliberately. The list also carries entries a broader
|
|
* glob already covers — `app/settings/system/storage/**` sits inside
|
|
* `app/settings/system/**`, `system-page-shell.tsx` inside
|
|
* `components/settings/system/*` — and #2202 kept those on purpose, as the
|
|
* record of which PR migrated which half of a group. That redundancy is
|
|
* intentional and must keep passing. Do not turn this into a subsumption check;
|
|
* it would fight a decision the list makes knowingly.
|
|
*/
|
|
export function duplicateEntries(entries) {
|
|
return [...new Set(entries.filter((entry, index) => entries.indexOf(entry) !== index))];
|
|
}
|
|
|
|
/**
|
|
* Entries that match no file at all — listed, but guarding nothing.
|
|
*
|
|
* Both other checks ask what happened to an entry BETWEEN two states: the
|
|
* removal check inspects entries that left the array, and `duplicateEntries`
|
|
* compares the array with itself. Neither asks the prior question — does this
|
|
* entry match anything? An entry that matched nothing on the day it was added
|
|
* never leaves, is not a duplicate, and passes `pnpm lint` trivially because the
|
|
* rule simply has no files to apply it to. The list grows, the ratchet reports
|
|
* "N added", and the path is unguarded. It is the allowlist's own version of a
|
|
* pass it had not earned: green everywhere, protecting nothing.
|
|
*
|
|
* A dynamic route is how you hit it, because Next-style directories are named in
|
|
* glob syntax and `[id]` is a CHARACTER CLASS matching a single `i` or `d`:
|
|
*
|
|
* app/settings/workspace/[id]/automations/**\/*.tsx -> 0 files
|
|
* app/settings/workspace/[[]id[]]/automations/**\/*.tsx -> 4 files
|
|
*
|
|
* #2247 shipped the unescaped form and review caught it; escaping it turned a
|
|
* hardcoded literal in that route from 0 lint errors into 1. Three older entries
|
|
* had the same defect. This makes the class impossible rather than the instances
|
|
* fixed.
|
|
*
|
|
* `resolve` is injected so this is testable against a fixture: the calling
|
|
* script resolves this repository's own paths from fixed locations and cannot be
|
|
* pointed elsewhere.
|
|
*/
|
|
export function unmatchedEntries(entries, resolve) {
|
|
return entries.filter((entry) => resolve(entry).length === 0);
|
|
}
|
|
|
|
/** An entry is a glob only if it carries glob metacharacters. */
|
|
const GLOB_METACHARACTERS = /[*?[\]{}]/;
|
|
|
|
function isFile(fsImpl, fullPath) {
|
|
try {
|
|
return fsImpl.statSync(fullPath).isFile();
|
|
} catch {
|
|
// Missing, or a broken symlink — either way it selects nothing.
|
|
return false;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* The files an allowlist entry currently matches, rooted at `cwd`.
|
|
*
|
|
* Exported so the removal check, the unmatched check and the tests share ONE
|
|
* definition of "this entry has files" — they used to state it separately, and
|
|
* two copies of a rule about glob semantics is how they drift apart. `fsImpl` is
|
|
* a seam for tests; it takes `node:fs` in the script.
|
|
*
|
|
* **Directories do not count, and that is the whole subtlety.** ESLint flat
|
|
* config matches `files` patterns against FILE paths, so an entry of
|
|
* `components/foo` selects nothing — it does not stand for `components/foo/**`.
|
|
* `existsSync` would happily confirm the directory exists and report the entry as
|
|
* live, which would let exactly the born-dead entry this module exists to catch
|
|
* through under a different spelling. Both branches are therefore filtered by
|
|
* `isFile`: `globSync` returns directories too (`components/automations/*`
|
|
* matches `.../trigger-configs`), so it is not only the literal case.
|
|
*/
|
|
export function filesForEntry(entry, { cwd, fsImpl }) {
|
|
const candidates = GLOB_METACHARACTERS.test(entry) ? fsImpl.globSync(entry, { cwd }) : [entry];
|
|
return candidates.filter((candidate) => isFile(fsImpl, `${cwd}/${candidate}`));
|
|
}
|