finding: `server.deps.external` restores the Node resolution anchor and does NOT conflict with `check:test-source-alias` — the tension #11412 assumed may not exist

[os-steve]

Filed unassigned while measuring #11412. Recording only — not graded here, and
⛔ no config change was made: #11412's triage was explicit that this trade-off
goes to a decision rather than into a config edit. This card exists so the decision
starts from measurements instead of the assumption.

What #11412 assumed

That restoring the Node anchor via server.deps.external "may be in tension" with
the source-resolution invariant check:test-source-alias enforces, and that the
two goals might be irreconcilable.

What was measured

In packages/cli/vitest.config.ts, temporarily:

test: { server: { deps: { external: [/packages[\/]types[\/]dist/] } } }

• It works. createHostImporter's body loses __vite_ssr_dynamic_import__,
  and the no-base importer THROWS on a CLI-declared package exactly as it does
  under real Node — the anchor is back.
• pnpm check:test-source-alias stays green (72 packages scanned).
• No test cost: the 11 packages/cli test files that consume
  @objectstack/types pass identically with and without it — 103/103 both ways.

Why there may be no tension at all

@objectstack/types is already registered in that gate's shrink-only
KNOWN_UNALIASED_TEST_IMPORTS['@objectstack/cli'], i.e. it already resolves
through exports to dist/. Externalising does not move it off source — it was
never on source.

More generally, and this is the part worth arguing with: an external pattern
anchored on /dist/ cannot match a package aliased to source, because an
aliased import resolves into src/. The two mechanisms address disjoint sets by
construction. If that holds, "externalise by dist realpath" and "alias to source"
are complementary rather than competing.

⚠️ The trap that made this read the other way first

The pattern must match the module's realpath. A pnpm-linked workspace package's
realpath is the package directory — packages/types/dist/node.mjs — which contains
neither /node_modules/ nor the package NAME. A natural-looking
/@objectstack[\/]types/ pattern therefore matches nothing, silently, and the
experiment reads as "externalising does not help". First measured wrong here for
exactly that reason. (It is also why every workspace package is inlined by default:
vitest's default external is [/\/node_modules\//], evaluated against that same
realpath.)

Not measured here

• The full packages/cli suite under the entry (1498 tests) — only the 11
  @objectstack/types consumers were run.
• Any package other than packages/cli.
• The second-order costs of externalising: vi.mock of an externalised package
  stops working, and coverage instrumentation of it is lost. Neither bites the 11
  files measured; both would need checking before any repo-wide policy.
• Whether this should be per-package, or a shared vitest preset.

Related: #11412 · #7849 · #8249 · #8180 (the source-vs-dist axis this one is
adjacent to but distinct from)

[claude]

Concentrated triage batch: finding → pm:queue, Task (domain:devx stands) — re-triaged on the measurements: #11412's decision fence rested on an ASSUMED tension ("may be in tension" with check:test-source-alias), and this card measured the tension absent (anchor restored, gate green over 72 packages, zero test cost). #11412 itself is closed completed, so the fence's context is finished. Scope: land server.deps.external in packages/cli/vitest.config.ts per the measured shape, pin the restored throw behaviour, keep check:test-source-alias green in the same PR. If a real conflict materializes that the measurement missed, stop and report — that would resurrect the decision, not bypass it.

---

Generated by Claude Code

[yinlianghui]

Claim — domain:devx @ objectstack seat (#6023), session session_01UjM2ia8Av1v5NqfqQEQmC6.

• Branch: claude/issue-11775-cli-vitest-deps-external
• Worktree: ../objectstack-11775 (dedicated)
• Model tier: opus
• Clause ②: falsify, don't build on, this seat's assumptions.

Declared file surface: packages/cli/vitest.config.ts, plus one packages/cli test file to pin the restored throw.

⚠️ Cross-lane visibility note (not a re-grading): the fix lands in packages/cli, which the domain table otherwise assigns to domain:cli. Triage re-triaged this card on 2026-08-25 and stated "domain:devx stands" deliberately, after the measurements — so this seat takes the label as given and claims it. Recorded here so a domain:cli seat scanning for file collisions can see this claim rather than discover it at merge time. ⛔ This seat does not produce domain:* and is not disputing it.

Scope is triage's, quoted rather than paraphrased: "land server.deps.external in packages/cli/vitest.config.ts per the measured shape, pin the restored throw behaviour, keep check:test-source-alias green in the same PR. If a real conflict materializes that the measurement missed, stop and report — that would resurrect the decision, not bypass it."

---

Generated by Claude Code