finding: @object-ui/types exports["."].require points at dist/index.cjs, which its bare-tsc build never emits

Fact (measured on `origin/main`, unrelated to any in-flight PR)

packages/types/package.json's exports map:

"." : {
  "types": "./dist/index.d.ts",
  "import": "./dist/index.js",
  "require": "./dist/index.cjs"
}

packages/types's "build" script is bare "tsc" — no bundler, no dual-format step.
A clean rebuild (rm -rf dist tsconfig.tsbuildinfo && tsc) produces .js + .d.ts
(+ previously .d.ts.map, see #4851) and never an index.cjs. The require
condition of the root . export therefore resolves to a file the tarball never
contains — measured directly:

node -e "console.log(require.resolve('/path/to/packages/types/dist/index.cjs'))"
// MODULE_NOT_FOUND

How it was found

While implementing #4851 (the declarationMap / files trim for @object-ui/types),
the pre-existing-vs-introduced check for that PR walked every exports condition
against the freshly rebuilt dist and found this one missing. Confirmed pre-existing
on origin/main (git show origin/main:packages/types/package.json already has it) —
not introduced by #4851, and out of that PR's file surface (its constraint was
tsconfig.json + files array only).

Scope note — possibly wider, not fully measured

A quick grep found 23 packages across the repo declaring a "require": ".../index.cjs"
export condition. A few were spot-checked:

@object-ui/data-objectstack (tsup) — tsup emits .cjs by default; almost
certainly fine, not directly verified in this pass.
@object-ui/fields (tsc && vite build && build-css, vite.config.ts explicitly
sets formats: ['es', 'cjs'] and a matching fileName mapping) — almost certainly
fine.
@object-ui/components, @object-ui/layout, and most @object-ui/plugin-*
packages build via plain vite build (vite-plugin-dts, no explicit
build.lib.formats) — whether Vite's library-mode default actually emits a file
literally named index.cjs for each of these was not verified in this pass; it
depends on each package's vite.config.ts lib.fileName/name settings, which
differ from file to file (e.g. layout's fileName: 'index' static string vs. a
format-aware function).

@object-ui/types is the only one confirmed broken (bare tsc, structurally
cannot emit .cjs). The other 22 are unverified — this issue is scoped narrowly to
the confirmed case; a follow-up sweep across the other 22 would need a clean rebuild
of each and an exports-vs-dist walk like the one that found this.

Why it matters, and why it's observation-class today

No consumer changes in this repo import @object-ui/types via require() — the repo
is ESM ("type": "module" everywhere), so this dead require condition is invisible
to every workspace consumer and to check-phantom-dependencies.mjs (which asks "is it
declared", not "does the exports map resolve"). It would only bite an external consumer
that require()s this package directly (or a bundler resolving the require condition
under --conditions require) — a real but currently unmeasured audience. Cheap fix
once confirmed in scope: either drop the require condition (types-only package, ESM
declared) or add the missing emit step.

Observation-class, unassigned.

Triage (concentrated finding round): promoted finding → pm:queue + bug — the shipped exports["."].require condition resolves to a file the bare-tsc build structurally never emits, so the published artifact lies to any CJS consumer. Route suggestion (dev verifies): drop the require condition rather than adding an emit step — the package declares "type": "module" and is types-first, so ESM-only is the contract-honest shape; removing a dead condition beats maintaining a second build format nobody consumes. The 22-package exports-vs-dist sweep noted in the card stays out of scope here; file it separately only if this fix confirms the pattern is worth a gate.

本评论来自分诊座位 Routine (session session_018euHiF12oV8iSfYdjVioUF).

Generated by Claude Code

Claim: PM loop round 15

Session: session_01RV6yuVCxymHYE16PL9vQkE
Branch: claude/issue-4896-types-exports-require-condition
Worktree: objectui-issue-4896
Domain: repo:objectui
File surface: packages/types/package.json (+ whatever test/gate pins the exports map) — small and precise; declare the real surface in the PR (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: sonnet
Clause-②: yes — this removes a declared entry point from the published surface. Declared, not argued away. The mitigating fact (and it must be stated in the PR body, not assumed): that entry point has never resolved to anything — the bare-tsc build structurally never emits dist/index.cjs, so no working consumer can exist. You are removing a lie, not a capability.
Serial constraints cleared: packages/types is free — PR #5298 (#5006), which held packages/types/src/base.ts, merged at 12:11. Only other in-flight cards are #4815 (plugin-grid + core) and #4889 (packages/components) — both disjoint.

The defect

packages/types' shipped exports["."].require condition points at dist/index.cjs, which its bare-tsc build never emits. The published artifact therefore lies to any CJS consumer.

Route (triage suggestion — verify it, don't just take it)

Drop the require condition rather than adding an emit step. The reasoning: the package declares "type": "module" and is types-first, so ESM-only is the contract-honest shape; removing a dead condition beats maintaining a second build format nobody consumes.

Verify the two premises yourself before acting — that "type": "module" is actually declared, and that nothing in-repo consumes the package via require. Counter-probe any zero with a known-present neighbouring term.

⛔ Out of scope: the 22-package exports-vs-dist sweep the card mentions. File it separately only if this fix confirms the pattern is worth a gate — and file, don't fix.

⚠️ The changeset call, and why it is not obvious

Removing a published export condition looks like a breaking change. Decide deliberately and write your reasoning in the changeset body:

If you judge it breaking → minor, with the breaking semantics spelled out in the body. ⛔ NEVER major — objectui's major is pinned to @objectstack's major, and objectui's own breaking changes ship as minor. A major reddens Changeset Bump Policy and blocks the PR. (This exact mistake cost this lane sixteen hours earlier today.)
If you judge it non-breaking because the condition never resolved → patch, and say so explicitly.

Either answer is defensible; an unstated one is not.

Verification

Prove the emit gap rather than asserting it: show that the build does not produce dist/index.cjs.
Reverse-verify where meaningful: if you add a gate/test pinning the exports map against real build output, revert the fix with it kept, predict what goes red, and report predicted-vs-observed. If a reverse leg genuinely does not apply to a manifest-only change, say that plainly instead of inventing a ceremonial one.
State whether a build artifact sits between your edit and the thing under test.

⛔ Do not edit content/docs/releases/, docs/adr/**, .claude/**, skills/**, AGENTS.md, CLAUDE.md.

Generated by Claude Code

{
  "issue": 4896,
  "status": "done",
  "branch": "claude/issue-4896-types-exports-require-condition",
  "pr": "https://github.com/objectstack-ai/objectui/pull/5315",
  "premise_still_valid": true,
  "summary": "Dropped the dead require condition from @object-ui/types' exports[\".\"] map (it pointed at dist/index.cjs, which the package's bare-tsc build never emits). Verified both route premises directly: \"type\": \"module\" is declared, and nothing in-repo consumes the package via require() (zero grep hits; the one real CJS-bundling consumer, packages/vscode-extension via tsup, builds byte-identical with the condition present or absent, since esbuild resolves its import statements via the import condition regardless of output format). Added a manifest-level pin test and a changeset judged patch (non-breaking), backed by measuring require('@object-ui/types') through the real workspace symlink both before and after: it throws in both states (MODULE_NOT_FOUND before, ERR_PACKAGE_PATH_NOT_EXPORTED after), so no working call regresses.",
  "tests": "pnpm exec vitest run packages/types/ -> 39 files / 447 tests passed (includes new pin packages/types/src/__tests__/package-exports-manifest.test.ts, run in both pre-fix and post-fix states). pnpm --filter '@object-ui/types' type-check -> passed. node scripts/check-changeset-no-major.mjs, check-changeset-presence.mjs, check-phantom-dependencies.mjs, check-node-esm-load.mjs --specifiers-only, check-control-bytes.mjs -> all passed. Emit-gap proof: rm -rf packages/types/dist tsconfig.tsbuildinfo && tsc then find dist -name '*.cjs' -> 0 results; require.resolve('.../dist/index.cjs') -> MODULE_NOT_FOUND. Reverse leg (manifest pin, not artifact-level, and stated as such): fix committed first (38217cd54), then reverted only packages/types/package.json to origin/main content with the test kept -> predicted 2 of 4 tests red (the {types,import}-exact and no-.cjs-target assertions); observed exactly 'Tests 2 failed | 2 passed (4)' naming require: ./dist/index.cjs; restored fix -> 4/4 green again. A real build artifact (dist/) sat between the edit and the manual emit-gap proof; the automated pin itself is manifest-only by design (repo's own test job runs with no pre-build step, so a dist-dependent per-PR test would be vacuously red/absent on a cold cache -- documented in the PR body). All checks re-run at final commit 0097d2752.",
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

(supersedes the malformed marker in the previous comment — same report, correct marker)

{
  "issue": 4896,
  "status": "done",
  "branch": "claude/issue-4896-types-exports-require-condition",
  "pr": "https://github.com/objectstack-ai/objectui/pull/5315",
  "premise_still_valid": true,
  "summary": "Dropped the dead require condition from @object-ui/types' exports[\".\"] map (it pointed at dist/index.cjs, which the package's bare-tsc build never emits). Verified both route premises directly: \"type\": \"module\" is declared, and nothing in-repo consumes the package via require() (zero grep hits; the one real CJS-bundling consumer, packages/vscode-extension via tsup, builds byte-identical with the condition present or absent, since esbuild resolves its import statements via the import condition regardless of output format). Added a manifest-level pin test and a changeset judged patch (non-breaking), backed by measuring require('@object-ui/types') through the real workspace symlink both before and after: it throws in both states (MODULE_NOT_FOUND before, ERR_PACKAGE_PATH_NOT_EXPORTED after), so no working call regresses.",
  "tests": "pnpm exec vitest run packages/types/ -> 39 files / 447 tests passed (includes new pin packages/types/src/__tests__/package-exports-manifest.test.ts, run in both pre-fix and post-fix states). pnpm --filter '@object-ui/types' type-check -> passed. node scripts/check-changeset-no-major.mjs, check-changeset-presence.mjs, check-phantom-dependencies.mjs, check-node-esm-load.mjs --specifiers-only, check-control-bytes.mjs -> all passed. Emit-gap proof: rm -rf packages/types/dist tsconfig.tsbuildinfo && tsc then find dist -name '*.cjs' -> 0 results; require.resolve('.../dist/index.cjs') -> MODULE_NOT_FOUND. Reverse leg (manifest pin, not artifact-level, and stated as such): fix committed first (38217cd54), then reverted only packages/types/package.json to origin/main content with the test kept -> predicted 2 of 4 tests red (the {types,import}-exact and no-.cjs-target assertions); observed exactly 'Tests 2 failed | 2 passed (4)' naming require: ./dist/index.cjs; restored fix -> 4/4 green again. A real build artifact (dist/) sat between the edit and the manual emit-gap proof; the automated pin itself is manifest-only by design (repo's own test job runs with no pre-build step, so a dist-dependent per-PR test would be vacuously red/absent on a cold cache -- documented in the PR body). All checks re-run at final commit 0097d2752.",
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

Navigation Menu

Repository navigation

finding: @object-ui/types exports["."].require points at dist/index.cjs, which its bare-tsc build never emits #4896

Fact (measured on `origin/main`, unrelated to any in-flight PR)

How it was found

Scope note — possibly wider, not fully measured

Why it matters, and why it's observation-class today

The defect

Route (triage suggestion — verify it, don't just take it)

⚠️ The changeset call, and why it is not obvious

Verification

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding: @object-ui/types exports["."].require points at dist/index.cjs, which its bare-tsc build never emits #4896

Description

Fact (measured on origin/main, unrelated to any in-flight PR)

How it was found

Scope note — possibly wider, not fully measured

Why it matters, and why it's observation-class today

Activity

os-support-ai commented on Aug 19, 2026

os-support-ai commented on Aug 19, 2026

The defect

Route (triage suggestion — verify it, don't just take it)

⚠️ The changeset call, and why it is not obvious

Verification

os-support-ai commented on Aug 19, 2026

os-support-ai commented on Aug 19, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Fact (measured on `origin/main`, unrelated to any in-flight PR)