Repository navigation
chore(renovate): hold TypeScript below 7 - #3772
Conversation
typescript-eslint reads `ts.Extension.Cjs` while building its extra-file-extension set. TypeScript 7's native rewrite no longer exposes that legacy API surface, so type-aware linting crashes with a TypeError rather than degrading to a warning. Every typescript-eslint release through 8.68.0 still peers `typescript >=4.8.4 <6.1.0`, and the upstream report (typescript-eslint#12518) is closed as not planned. Bumping @bfra.me/eslint-config does not help — it carries 8.67.0, which is capped the same way. Nothing in this repo needs migrating when support lands: the resolved tsconfig already uses moduleResolution bundler and target ES2022, with no baseUrl, and scripts run on Node's strip-only loader rather than this TypeScript.
fro-bot
left a comment
There was a problem hiding this comment.
This is a targeted quarantine of a version bump that can't survive contact with the toolchain. The comment does the heavy lifting: it names the exact failure surface (ts.Extension.Cjs resolving undefined), cites the upstream (typescript-eslint#12518), records the peer range that gates the fix (>=4.8.4 <6.1.0), and specifies the signal to lift the hold. That's a load-bearing comment, not decoration — the next person who wonders why TS is frozen gets the whole map.
The constraint is mechanically sound. Current pin is typescript: 6.0.3; allowedVersions: '<7' filters 7.x candidates while 6.x updates still flow, and the additional matchPackageNames: ['typescript'] rule merges additively with the existing two — no collision. Scope is a clean +10/-0 in one config file. No workflow, permission, secret, or injection surface touched.
Verdict: PASS
Blocking issues
None.
Non-blocking concerns
.github/renovate.json5:24— The hold is<7, but the typescript-eslint peer actually caps at<6.1.0. A<7ceiling still lets Renovate surface6.1.x+ candidates that the peer would reject. In practice this is inert: patch-only is enabled fortypescript(line 8) and the pin is6.0.3, so Renovate only offers6.0.xpatches. If you want the hold to mirror the real constraint exactly,<6.1.0is the tighter boundary — but<7reads cleaner as intent and the patch-only rule already neutralizes the gap. Author's call.
Missing tests
None. This is declarative Renovate config, not executable code — there's no behavior to unit-test. JSON5 validity is CI's job, and the rule's effect is observable only against live Renovate resolution.
Risk assessment: LOW
Single-file config change that narrows an update range. Failure mode is bounded to "Renovate proposes a TS version it shouldn't," which the constraint exists to prevent. No automation, credential, or supply-chain surface altered. Worst case on a bad constraint is a rejected PR, not a broken pipeline.
Run Summary
| Field | Value |
|---|---|
| Event | pull_request |
| Repository | fro-bot/.github |
| Run ID | 33111248639 |
| Cache | hit |
| Session | ses_fbb2f81cdffe6J4XWSujeQmjSH |
typescript-eslint peers `typescript >=4.8.4 <6.1.0`, so 6.1.x breaks type-aware linting for the same reason 7.x does. Capping at <7 would have admitted it: no other rule here restricts TypeScript to patch-only updates, so a minor bump would flow normally.
|
Tightened the ceiling to The concern was right, though the reason it's not inert is different from the mitigation reasoning. The existing rule { matchUpdateTypes: ['patch'], matchPackageNames: ['!python', '!typescript'], enabled: false }disables patch updates for everything except python and typescript — it exempts TypeScript from the patch-disable rather than restricting it to patch. Minor updates aren't governed by that rule for any package, so a Mirroring the peer range is the right boundary. The comment now says why |
fro-bot
left a comment
There was a problem hiding this comment.
Re-review after 5cb73db. The ceiling moved from <7 to <6.1.0, and you were right to correct my reasoning on the way through.
My prior note called the <7 gap "inert" because typescript looked patch-only. That was a misread. The rule at .github/renovate.json5:8 — matchUpdateTypes: ['patch'], matchPackageNames: ['!python', '!typescript'], enabled: false — exempts typescript from the patch-disable; it doesn't restrict it to patch. Minor updates aren't governed by that rule for any package, so a 6.0.3 → 6.1.0 bump would have sailed through a <7 ceiling and detonated type-aware linting on the exact same ts.Extension.Cjs lookup as 7.x. The gap wasn't neutralized — it was live. <6.1.0 mirroring the peer range is the correct boundary, not just the tidier one.
The comment now carries that reasoning: it states why <6.1.0 and not <7, which stops the next person from loosening it back on the assumption that only the major is dangerous. Load-bearing again.
Verdict: PASS
Blocking issues
None.
Non-blocking concerns
None. The prior concern was resolved and the underlying justification is now tighter than my original framing.
Missing tests
None. Declarative Renovate config — no executable behavior to unit-test. JSON5 validity is CI's gate; the rule's effect is only observable against live Renovate resolution.
Risk assessment: LOW
Single-file config change narrowing an update range to match a documented upstream peer ceiling. Failure mode is bounded to Renovate proposing a TS version it shouldn't — which is exactly what this now prevents, including the minor-version path the previous ceiling left open. No workflow, permission, secret, or supply-chain surface touched.
Run Summary
| Field | Value |
|---|---|
| Event | pull_request |
| Repository | fro-bot/.github |
| Run ID | 33116918076 |
| Cache | hit |
| Session | ses_fbb2f81cdffe6J4XWSujeQmjSH |
fro-bot
left a comment
There was a problem hiding this comment.
Re-review after the 51269b0 merge of main into the branch. The diff against main is byte-identical to what I approved in the prior run: the <6.1.0 ceiling holds, the load-bearing comment survived the merge intact, and no conflict residue landed in the file. This is a housekeeping merge to keep the branch current, not a change to the logic.
The substance was already settled: the hold mirrors the typescript-eslint peer range (>=4.8.4 <6.1.0) rather than the looser <7, closing the minor-version path (6.0.3 → 6.1.0) that would have detonated type-aware linting on the same ts.Extension.Cjs lookup as 7.x. The comment states why <6.1.0 and not <7, so the boundary won't get loosened back by someone who assumes only the major is dangerous.
Verdict: PASS
Blocking issues
None.
Non-blocking concerns
None. Nothing changed since the prior approval except a clean merge from main.
Missing tests
None. Declarative Renovate config — no executable behavior to unit-test. JSON5 validity is CI's gate; the rule's effect is only observable against live Renovate resolution.
Risk assessment: LOW
Single-file config change narrowing an update range to a documented upstream peer ceiling, unchanged across the merge. No workflow, permission, secret, or supply-chain surface touched. Worst case on a bad constraint is a rejected Renovate PR, not a broken pipeline.
Run Summary
| Field | Value |
|---|---|
| Event | pull_request |
| Repository | fro-bot/.github |
| Run ID | 33117949085 |
| Cache | hit |
| Session | ses_fbb2f81cdffe6J4XWSujeQmjSH |
Renovate keeps proposing TypeScript 7 and it can't be merged. I closed #3713; this stops it coming back every cycle.
typescript-eslint reads
ts.Extension.Cjswhile building its extra-file-extension set. TypeScript 7's native rewrite doesn't expose that legacy API surface, so the lookup resolvesundefinedand type-aware linting crashes outright rather than degrading to a warning:Every typescript-eslint release through 8.68.0 still declares
typescript >=4.8.4 <6.1.0, and the upstream report (typescript-eslint#12518) is closed as not planned. Bumping@bfra.me/eslint-configdoesn't route around it either — 0.52.1 carries typescript-eslint 8.67.0, capped the same way. Microsoft's own guidance is that the native implementation doesn't preserve the Strada compiler API for tooling, and points at side-by-side installs rather than a shim.Nothing here needs migrating when support lands. The resolved tsconfig already uses
moduleResolution: "bundler"andtarget: "ES2022"with nobaseUrl, so none of TS7's removals apply, andscripts/runs on Node's strip-only loader rather than this TypeScript — the pin only affectscheck-typesandlint.The comment on the rule names the exact signal for lifting it:
@typescript-eslint/parserpublishing a peer range that permits 7.x.