Skip to content

Release 0.2.1: the CLI starts again and database stacks build - #3

Merged
amagdy46 merged 20 commits into
mainfrom
release-0.2.1
Nov 19, 2026
Merged

amagdy46 merged 20 commits into
mainfrom
release-0.2.1

Conversation

@amagdy46

@amagdy46 amagdy46 commented Nov 18, 2026 •

Copy link
Copy Markdown
Owner

npm create efforge crashes on startup today, and most generated projects fail pnpm build. This fixes both, fixes the other bugs a new user hit, and adds the CI guards that would have caught it.

What was broken

  • The CLI crashed before printing anything, under npm and pnpm dlx. 0.2.0 pinned @effect/platform-node@4.0.5, which depends on @effect/platform-node-shared: ^4.0.5. That range now resolves to 4.1.4, which needs Effect 4.1.
  • Generated projects with a database, or with realtime and no database, failed pnpm build. drizzle-orm and effect-cf pulled Effect 4.1 SQL packages in next to Effect 4.0.5.
  • CI's e2e matrix never ran a stack. Vitest joins test names with >, so the -t filter matched nothing and every e2e job passed with 17 tests skipped. CI also had no scheduled run.
  • Smaller bugs: the realtime demo was missing from database and auth dashboards; add left a pnpm-lock.yaml behind and then warned about it; add generated a route tree on a never-installed project; a mistyped flag printed 49 lines of help, also under --json; docs and errors still said "v0.1".

What changed

  • The CLI is one bundled file with no runtime dependencies (ADR 0004). Nothing is resolved on the user's machine, under any package manager, so a dependency release cannot break an already published version again. --help went from 682 ms to 265 ms.
  • Generated projects hold every Effect 4 package at one version with overrides in pnpm-workspace.yaml, and pin pnpm 12.21.0 (ADR 0005). Libraries that depend on Effect 3 keep their own copy.
  • add repairs projects created by 0.2.0: it writes the overrides and raises packageManager when the project's pnpm is too old to honour them. The changelog has the manual repair.
  • Pins bumped: Effect 4.1.4, effect-cf 0.59.0, better-auth 1.7.25, Vite+ 1.7.0, @effect/tsgo 0.70.0, Alchemy beta.87 (provider moved to @alchemy.run/cloudflare), the official Drizzle 1.0.0-rc.5. Vitest stays at 5.1.0 on purpose, to match the copy Vite+ bundles.
  • CI: the e2e filter is fixed and each e2e job fails unless exactly one test passed; a daily and on-demand run without the pnpm cache; a packed-tarball smoke job; a job for the add e2e case.
  • release.yml packs once, smoke-tests that tarball and publishes the same file, still with trusted publishing and provenance.
  • Version 0.2.1 and its changelog section.

How it was verified

Locally, at the head commit:

  • Repo gate (typecheck, lint, format, build): 259 tests pass.
  • pnpm smoke:pack: the tarball installed with npm in an empty directory with no lockfile, run through npm, npx and pnpm dlx; creates the default stack, builds it, adds email, builds again.
  • The whole e2e matrix: 15 stacks plus the add case pass. The browser flow passes against Postgres 17.
  • Every audit repro, against the packed tarball. A project generated by the v0.2.0 tag, then add email with this branch: pnpm build passes and pnpm peers check is clean.
  • The regression case itself: the old 4.0.5 pins plus the new overrides install nothing from 4.1.x, and the database stacks build.
  • Two fresh-context reviews of the two source branches, their findings fixed, and a re-review of the fixes that re-ran each repro on projects generated from the v0.2.0 tag: all addressed.

On GitHub: this pull request's CI is the first time the e2e matrix has really run. Its first run passed 20 of 21 jobs and caught one more real bug: with CI set, pnpm freezes the lockfile, so the install after efforge add failed with ERR_PNPM_OUTDATED_LOCKFILE. add now installs with --no-frozen-lockfile, and all 21 checks pass.

Still only provable after merge: the daily schedule, and publishing through the changed release workflow.

To release

  1. Merge to main.
  2. git tag v0.2.1 && git push origin v0.2.1. release.yml smoke-tests the tarball and publishes it.

0.2.0 crashed on start because @effect/platform-node's own range resolved
@effect/platform-node-shared to 4.1.4. A bundled CLI resolves nothing on the
user's machine. pnpm smoke:pack installs the packed tarball with npm in an
empty directory and runs it; release.yml runs it before publishing.
drizzle-orm and effect-cf peer on @effect/sql-* with open ranges, so pnpm
installed 4.1.4 drivers next to effect 4.0.5 and every database stack failed
to build. The generated pnpm-workspace.yaml now overrides each family
released with effect to the effect pin; add writes the overrides into older
projects at their own effect version. Wildcard overrides need pnpm 12.17,
so generated projects pin pnpm 12.21.0. The e2e harness checks the real
lockfile and pnpm peers check after every install.
…ilter

Vitest 5.1 joins describe and test names with " > ", so -t "stacks build
<case>$" matches nothing and skips every case.
effect and @effect/* 4.1.4, effect-cf 0.59.0, alchemy beta.87,
better-auth 1.7.25, vite-plus 1.7.0, @effect/tsgo 0.70.0, oxlint 1.94.0,
drizzle-orm and drizzle-kit to the official 1.0.0-rc.5, and the rest of
bump.ts's list. vitest stays at 5.1.0, the copy vite-plus 1.7.0 bundles.

vite-plus 1.7 runs type-aware lint on its bundled tsrs, so the tsgolint
that effect-tsgo patches went unused and lint failed on an unknown
effecttsgo rule. The lint and check scripts now run through
effect-tsgo vp. bump.ts checks pins against tsgo's supported table and
vite-plus's bundle. The browser extra's lockfile and image tag follow
playwright 1.65.1.
alchemy 2.0.0-beta.87 moved the Cloudflare provider out of the alchemy
package (alchemy/Cloudflare no longer exists) into @alchemy.run/cloudflare.
The e2e peer check now fails on required or Effect peers only, and names
the one upstream mismatch it tolerates (alchemy's runtime pulls unenv rc.25
next to @cloudflare/unenv-preset's exact rc.24 peer).
# Conflicts:
#	CONTRIBUTING.md
#	docs/limits.md
#	scripts/smoke-pack.ts
… add; release publishes the tarball it tested; CI runs the add e2e case
… Effect overrides, scope the effect override to Effect 4, refresh an unedited dashboard so the realtime demo shows
…t checks git status and the warning, a test that --json --log-level=nope stays one JSON error
@amagdy46
amagdy46 marked this pull request as ready for review November 18, 2026 17:42
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 18, 2026 •

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-18T17:44:28.651612Z 3f4fa0f Draft marked ready
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@amagdy46
amagdy46 merged commit cd3f4e5 into main Nov 19, 2026
21 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant