fix(lint): clear latent dead stores and a dropped assertion

[frantuma]

Prepares the ground for the eslint 10 upgrade, and fixes eleven real findings on its own merits.

ESLint 10 enables no-useless-assignment and preserve-caught-error in its recommended set. Both flag genuine problems in this repository, so they are fixed here — separately and reviewably — rather than buried inside a dependency bump.

Every change in this PR is valid under the eslint 9 currently on main; nothing here depends on the upgrade.

Dead stores (7)

nodePath in the two pointer helpers was declared and seeded outside a loop that reassigns it to [] on every iteration, so it is now a const scoped to the loop body. conditionRes was seeded true and then assigned in both branches of the following if/else. The formatting service seeded result with textDocument.getText() and overwrote it with the prettier output as the first statement of the try, so the declaration moves to that assignment. getText assigned '' and then overwrote it in both branches of an if/else, which is now a single conditional assignment.

A swallowed cause (1)

The formatting service catches a prettier failure, builds a message from it, and rethrows:

throw new Error(error);          // before — the prettier error is lost
throw new Error(error, { cause: e });  // after

The message already interpolated e.message, but the original error object — with its stack and any structured detail — was discarded. This is the one change with a runtime effect: consumers catching the error now get .cause.

Unasserted results (4) — the more interesting half

All four were calls whose result was thrown away. They are not treated identically, because they are not the same thing.

test/detect.ts, two sites — a dropped assertion, now restored. The file parses each fixture twice, once without a content language and once with, and asserts both. Every block does this except the two apidom fixtures, which parse twice and assert only the second. Rather than guess, I measured what the first parse returns:

apidom.json: parse(doc, undefined) -> api.element = undefined
apidom.yaml: parse(doc, undefined) -> api.element = undefined

An apidom-format document resolves to no api element without a content language — which is presumably exactly why the assertion was never written. It is now stated:

assert.isUndefined(result.api, 'an apidom document needs a content language to resolve');

That is a real behaviour, and the neighbouring findNamespace assertions already confirm the same document resolves to namespace apidom without a content language and asyncapi with one.

test/openapi-json.ts and test/asyncapi-yaml.ts — deliberately left unasserted. Both keep the call and lose only the assignment. Asserting them would mean inventing expectations:

• openapi-json validates a spec commented // valid spec and discards the result before replacing the document. That spec actually produces 9 diagnostics, measured, not zero. Pinning nine unreviewed diagnostics into an assertion is not something this PR should do quietly.
• asyncapi-yaml sits under an open // TODO yaml errors not recovered? no result? with a commented-out expectation block. Asserting current behaviour there would contradict the open question.

Keeping the call preserves what it still does — catch a throw — while removing the dead store. Both remain fair game for a follow-up that decides what they should assert.

One knock-on: dropping the reassignment in asyncapi-yaml made its result const-eligible, so prefer-const applied.

Verification

check	eslint 9 (main)	eslint 10
npx eslint ./	0 errors, 14 warnings	0 errors, 14 warnings
npx tsc --noEmit	exit 0	—
npm test	400 passing	—

The eslint 10 column is the point of the PR, and was checked by resolving eslint 10 into the tree without committing it: no-useless-assignment and preserve-caught-error report no errors at all afterwards, down from 11. So the dependency bump that follows can land green.