Skip to content

Raise the core peer range of duxt-wiki to 0.13.0 #161

Closed

Description

@TitusKirch

Before you start

  • I searched existing issues and didn't find a duplicate
  • This is not a security vulnerability

What happened?

packages/duxt-wiki/package.json declares @kirchdev/duxt as a peer with >=0.4.0 <1. The wiki layer needs core features that first ship in @kirchdev/duxt 0.13.0: build contributions through duxt.build.ts, page types and the ./unstable export. @kirchdev/duxt 0.12.0 has no ./unstable export.

Expected behavior

Installing @kirchdev/duxt-wiki next to a core older than 0.13.0 produces a peer dependency warning instead of an install that looks fine and then does not work.

Steps to reproduce

  1. Install @kirchdev/[email protected] and @kirchdev/duxt-wiki in a site.
  2. Extend both layers.
  3. The package manager reports no peer conflict, although the core lacks what the layer needs.

Affected area / component

packages/duxt-wiki/package.json

Version & environment

@kirchdev/duxt-wiki 0.13.0, the first real release of the layer.

Anything else?

Raise the lower bound to >=0.13.0 <1. It stays a wide range rather than ^0.x, as for the other companion layers. Decided on 2026-11-09: this ships with the next release, not in the 0.13.0 release pull request (#160).

Activity

  1. added
    bugSomething isn't working
    needs triageNot yet triaged by a maintainer
    ai: readyScoped + approved for an AI agent to pick up
    ai: workingAn AI agent is implementing (incl. committed but not yet pushed)
    and removed
    needs triageNot yet triaged by a maintainer
    ai: readyScoped + approved for an AI agent to pick up
    on Nov 8, 2026
  2. self-assigned this
    on Nov 12, 2026
  3. added and removed
    ai: workingAn AI agent is implementing (incl. committed but not yet pushed)
    on Nov 12, 2026
  4. TitusKirch commented on Nov 12, 2026

    @TitusKirch
    MemberAuthor

    Implemented on dev in df000ed: the @kirchdev/duxt peer range of packages/duxt-wiki is now >=0.13.0 <1, as the body specifies. The lockfile needed no change.

    Checked against the repository before the change: everything the wiki layer imports from the core — the ./unstable exports including DuxtBuildContribution, sourcesForRoute and the useDuxt* composables — already exists in [email protected], and [email protected] has no ./unstable export.

    Test discipline: a manifest-only change, so no test was driven, and no local checks were run at the maintainer's request; the change ships through promotion PR #169, whose CI gate covers it.

  5. added
    ai: reviewingAn AI agent is reviewing the pushed work
    and removed on Nov 12, 2026
  6. TitusKirch commented on Nov 12, 2026

    @TitusKirch
    MemberAuthor

    AI review — needs human (round 1 of 3, commit df000ed)

    The change is correct and complete. It cannot be accepted yet, because the CI gate on this head is red for a reason outside this issue's commits.

    The diff (df000ed, the only commit referencing #161)

    packages/duxt-wiki/package.json changes its peer range from "@kirchdev/duxt": ">=0.4.0 <1" to ">=0.13.0 <1". That is exactly what the body asks for. The range stays wide and is not ^0.x, and nothing outside the request changed.

    What I verified independently

    • 0.13.0 is the right lower bound. [email protected]'s exports map has no ./unstable. [email protected] adds it, pointing at ./build/unstable.ts. The wiki imports DuxtBuildContribution, readContentCache, writeReportNotes, linkWarnings and walk from @kirchdev/duxt/unstable (module.ts, duxt.build.ts, infobox-build.ts). All five are exported at [email protected], and build/unstable.ts has not changed between that tag and dev. The other subpaths the wiki uses (sources-resolve, sections-resolve, sections-openapi) and the runtime symbols it calls (sourcesForRoute, useDuxtCollection, useDuxtConfig, useDuxtLink, useDuxtNavigation, useDuxtPageFocus, useDuxtPath, useDuxtSection, useDuxtSectionRow, useDuxtSiteUrl) already exist at v0.12.0. So ./unstable is the dependency that sets the bound.
    • The bound is not too low. After [email protected] the wiki gained three commits (03f0b00, 3cfbbda, and the dependency bump f5ef54e). None of them imports anything from the core that 0.13.0 lacks.
    • The lockfile needed no change. The wiki's importer entry records @kirchdev/duxt only as the workspace:* devDependency. The peer specifier does not appear in it. CI's pnpm install --frozen-lockfile on this head reported "Lockfile is up to date".
    • Nothing else in the body was required. No wiki docs page or README states the peer range. ADR-0011's >=0.4.0 <1 describes the search providers and already says "the lower bound being whatever the provider first needs". The typesense and meilisearch ranges are out of scope.

    Why this is not done

    Check source: the forge checks of promotion PR #169, whose head is df000edb. No local gate was run.

    • Passed: CodeQL (Gate and both Analyze jobs), Adapters (libsql, postgresql), Content packer (images).
    • CI / Lint & Format failed, and CI / Gate failed with it (run 43329518143). lint, format, tsc and typecheck:app passed. In turbo run test, 4 of 5 package tasks passed. @kirchdev/duxt#test failed on one test: tests/content-bundle-server.test.ts › "refreshes real Content, media, redirects, MCP and caches without restarting". It timed out at 180000 ms; the other 4790 tests passed.
    • The failure does not come from Raise the core peer range of duxt-wiki to 0.13.0 #161. That test runs the core's own content-bundle server, and a peer specifier in another package cannot reach it; the lockfile and install were unchanged. On the previous head, 286b2776, the same suite passed, and that run stopped later, at check:security. That suggests a timing-sensitive test under runner load rather than a regression, but nothing has confirmed it.
    • Because the gate stops at the first failed task, everything after test never ran on this head: check:policy, check:previews, check:security, build:app and the browser checks. So this head has no green gate, and "the PR chore: merge dev into main #169 gate covers it" does not hold yet.

    What a human needs to do

    Re-run the failed CI job on #169. If it comes back green, this issue can go to done with no code change. If content-bundle-server.test.ts times out again, track it as its own issue against the core's test, not against #161.

  7. added
    ai: needs humanEscalated - a human must review or decide
    ai: doneAI-reviewed and accepted
    and removed
    ai: reviewingAn AI agent is reviewing the pushed work
    ai: needs humanEscalated - a human must review or decide
    on Nov 12, 2026
  8. TitusKirch commented on Nov 12, 2026

    @TitusKirch
    MemberAuthor

    Accepted by the maintainer: the change is correct and complete, and the red gate on df000edb was a timeout in the core's content-bundle-server test that this peer range cannot reach. That test's own budget is fixed separately.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions