dshmarket 1.73.19: One Test Fix, and a Release Outside the Evening Slot (Nov 12, 2026)

The evening rhythm moved: per the npm registry (fetched 2026-11-12), 1.73.19 went up at 06:41:18 UTC on November 12, about fourteen hours after the previous evening’s 1.73.18 — dshmarket 1.73.18: Nine Changes After a Skipped Evening, Three First-Time Contributors (Nov 11, 2026) covered that release and its nine changes. The GitHub release page stamps 06:37:12 UTC, authored by the repo’s github-actions bot as before, and links the compare range v1.73.18…v1.73.19.

One change, test-only

Per the release notes, What’s Changed holds exactly one item — PR #1178, titled test: keep npm test off the network and green on Windows, by @BBleae — and the notes carry no New Contributors section. Per the PR record, it was opened on 2026-11-11 at 18:47 UTC, merged at 20:38 UTC against the nightly1014 branch, and the release followed just under ten hours later.

Per the PR body, both problems it fixes live in the unit lane, and nothing under src/ changes. The first is network reachability: the file-level beforeEach in tests/flows.spec.ts answered only the registry package-latest endpoints on npmjs.org and npmmirror.com and handed every other URL to the real fetch — the same body quotes the stub’s own comment, everything else goes to the real fetch — while vitest.config.ts describes the lane as no network. Under a preload that refuses and logs every non-loopback fetch, one run of the flows spec logged 11 outbound requests — 3 to registry.npmjs.org, 6 to github.com, 2 to raw.githubusercontent.com — and all 443 specs still passed without them. The fix turns the stub into a refuser: a request nothing answers gets TypeError('unit lane is offline: no stub answers <url>'), a new spec pins the lane offline, and the three places that wrap whatever fetch came before them forward into the refusing stub.

The second is a Windows failure in tests/pnpm-compat.spec.ts, the one failure in the check (windows-latest) job of the CI run for release 1.73.18: the spec expects the error message to name PNPM_CONFIG_STORE_DIR=, but on Windows environment names are case-insensitive, so stubbing the uppercase spelling and then pnpm_config_store_dir operates on the same variable — and after vi.unstubAllEnvs() restores it, the variable survives lowercase, the message correctly names the spelling process.env lists, and the assertion misses. The spec came in with #1173, per the body, which is the pnpm store reconnect advice release dshmarket 1.73.18: Nine Changes After a Skipped Evening, Three First-Time Contributors (Nov 11, 2026) covered as PR #1173. The fix asserts the variable under the name process.env actually lists — on Linux and macOS that is the same uppercase string as before — and makes the no-variable-may-be-named assertion above it case-insensitive so it cannot pass vacuously on Windows.

Cadence

Per the same 2026-11-12 registry fetch: twenty of the twenty-two calendar days from 10-22 to 11-12 carry at least one dshmarket release, the quiet ones staying 10-25 and 11-10. Within the 1.73.x line, the nineteen versions from 1.73.0 (2026-10-27) through 1.73.18 all carry npm stamps between 16:49 and 21:26 UTC, and 1.73.19’s 06:41:18 UTC stamp is the only one outside that span. The release page is still authored by the dsh-market repo’s github-actions bot rather than a human account.

One change, test-only

Cadence

More in this series