Skip to content

Merge main into preprod: Prisma 7 + mesh 2.0 migration + test-infra fixes#269

Open
QSchlegel wants to merge 13 commits into
preprodfrom
staging/preprod-prisma7-mesh2
Open

Merge main into preprod: Prisma 7 + mesh 2.0 migration + test-infra fixes#269
QSchlegel wants to merge 13 commits into
preprodfrom
staging/preprod-prisma7-mesh2

Conversation

@QSchlegel

@QSchlegel QSchlegel commented Jun 8, 2026

Copy link
Copy Markdown
Collaborator

Brings main into preprod and completes the two breaking dependency upgrades main introduced, plus the test-infra work needed to keep the suite green. This unblocks PR #229 (preprod → main), which currently can't merge due to conflicts.

All gates green

  • tsc --noEmit: 0 errors
  • test:ci: CJS 357 + ESM 48 passed, 0 failures
  • next build: success

What's included

Conflict resolution (16 files) — workflows, jest config, package.json, prisma schema, several test + component files. Notable calls: kept preprod's backwards-compat nostrKey, defensive txJson parse (#211), extracted buildStakingActionConfigs; combined async getTxBuilder (cost models) with useCslSerializer; dropped unused @hookform/resolvers + @jinglescode/nostr-chat-plugin.

Prisma 6 → 7 migration (main bumped the version but never migrated):

  • prisma.config.ts for CLI connection URLs; removed url/directUrl from schema
  • @prisma/adapter-pg driver wired into src/server/db.ts (Prisma 7 requires an adapter)
  • shared createPrismaClient() factory for 6 CI scenario scripts that constructed new PrismaClient() directly
  • fixed the @prisma/client/runtime/library Decimal import

Mesh react 2.0 (redesigned, incompatible wallet API): a useMeshWallet bridge sources a 1.9 IWallet from react 2.0's connection so all signing/tx code keeps the exact 1.9 API — deliberately avoiding the signData(payload, address) → signData(address, data) argument-swap footgun. Null-guards added at call sites.

Test infrastructure: the suite mixes jest.mock (CJS) and unstable_mockModule (ESM) tests, which the upgrades pushed past breaking under a single jest mode. Split into two configs from a shared base (jest.shared.mjs → jest.config.mjs CJS + jest.esm.config.mjs ESM), run via scripts/run-tests.mjs. Coverage is CJS-only (per-file threshold lives there; v8 coverage + native ESM crashes), and collectCoverageFrom excludes import-side-effect modules (src/server/**, src/lib/security/**) that polluted mocked tests. trpc/* DB integration tests stay in their own workflow.

Follow-ups / watch items

  • Runtime validation needed before promoting to prod: the mesh-2.0 bridge keeps the 1.9 signing API but the wallet object is re-sourced — exercise a real signing/voting flow. The Prisma 7 pg-adapter connection on Supabase + Vercel serverless should be smoke-tested (DIRECT_URL for migrations bypasses pgbouncer).
  • The bridge is a pragmatic interim; a full migration to the react-2.0 wallet API can follow later.

🤖 Generated with Claude Code

QSchlegel and others added 3 commits June 8, 2026 14:02
@QSchlegel @claude
Merge main into preprod: resolve conflicts + migrate Prisma 7 & mesh 2.0
3166bff 
Snapshot on staging branch for deeper test-infra testing. Captures the full
resolution of PR #229's 16 conflicts plus two breaking dependency upgrades
from main that required migration:

- Prisma 6->7: add prisma.config.ts, drop schema url/directUrl, wire the
  @prisma/adapter-pg driver in db.ts, fix Decimal import.
- mesh react 2.0: useMeshWallet bridge sources a 1.9 IWallet from react 2.0's
  connection so signing/tx code keeps the 1.9 API (avoids the signData arg
  swap). Null-guards added at call sites.

tsc is green (0 errors). Known WIP: jest suite has an ESM/CJS mock-mode
conflict exposed by the upgrades (jest.mock vs unstable_mockModule) - test
infra fix in progress.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
test-infra: complete Prisma 7 migration in CI scripts + fix test loading
6bb94db 
- Fix 6 CI-scenario `new PrismaClient()` sites that Prisma 7 rejects (no
  adapter): add scripts/ci/framework/prismaClient.ts factory and route all
  construction through it.
- Add @/env jest stub (src/__tests__/__mocks__/env.cjs) + moduleNameMapper so
  server modules load under jest without the ESM-only @t3-oss validator.
- Fix botAuth jsonwebtoken mock to expose `default` (esModuleInterop default
  import was undefined).
- Keep preprod's plain-jest invocation (revert the --experimental-vm-modules
  change, which broke jest.mock-based tests).

Bot gate now 128/132; full suite 373/421. Remaining failures are
unstable_mockModule (ESM) tests that don't apply under plain jest - mock-style
unification still pending.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
test-infra: split CJS/ESM jest projects so the full suite passes
7e082f0 
The suite mixes two mutually-exclusive jest module modes (jest.mock = CJS,
unstable_mockModule/import.meta/ESM-deps = ESM). The mesh 2.0 + Prisma 7
upgrades tripped the ESM-loading paths that preprod's older deps avoided, so
plain jest could no longer run both. Resolved by running two jest invocations
from a shared base (jest.shared.mjs):

- jest.config.mjs (CJS, plain jest): all tests except the 8 ESM files and the
  trpc/* DB integration tests (their own workflow).
- jest.esm.config.mjs (ESM, --experimental-vm-modules): the 8 unstable_mockModule
  / import.meta files.
- scripts/run-tests.mjs runs both and forwards args; test/test:ci/test:bot:*
  go through it. --runInBand makes the gate deterministic; --coverage is
  CJS-only (v8 coverage + native ESM crashes, and the per-file threshold lives
  in the CJS config).
- collectCoverageFrom excludes src/server/** and src/lib/security/** — jest
  force-loads uncovered files for coverage and their import-time side effects
  polluted mocked tests (freeUtxos).
- Fix signTransaction.test @meshsdk/core mock to export resolveStakeKeyHash
  (ESM static linking needs it).

Result: test:ci green (CJS 357 + ESM 48, 0 failures), tsc 0 errors.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@vercel

vercel Bot commented Jun 8, 2026 •
edited
Loading

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
multisig Ready Ready Preview, Comment Jun 8, 2026 4:16pm

Request Review

@QSchlegel @claude
fix(deps): restore nested @simplewebauthn/browser@9.0.1 in lock
c680f5e 
My npm 11 lock regen deduped @simplewebauthn/browser to 13.3.0, which doesn't
satisfy @auth/core's ^9.0.1 — CI (npm 10, node 20) rejected `npm ci` as out of
sync. Rebuilt the lock with npm 10 (--package-lock-only) to restore the nested
9.0.1 entry under @auth/prisma-adapter. Key versions unchanged.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
test(trpc): restore real timers for DB integration tests
2b7da9e 
The merged setup.ts (from main) freezes the clock (Date.now/new Date) for unit
determinism. tRPC integration tests hit a real Postgres and Prisma 7 sets
createdAt client-side, so a frozen clock gives identical timestamps and breaks
ordering assertions. Run test:trpc via a dedicated config that layers a
real-timers setup over the global one.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
fix(build): transpile @meshsdk/core-csl + whisky-evaluator for WASM
7fae786 
@meshsdk/core-csl 1.9 pulls whisky-evaluator, which ships a .wasm the Node
server runtime can't load when the package is externalized
(ERR_UNKNOWN_FILE_EXTENSION ".wasm" → createWallet 500). Add both to
transpilePackages so Next bundles them and applies its WebAssembly handling.

Candidate fix — needs the v1 smoke / Vercel deploy to confirm the WASM loads at
runtime (the local build still warns about the .wasm modules).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
ci(smoke): run v1 smoke against a production build, not next dev
578db8c 
The multisig-v1-smoke ran the app via `next dev`, which mis-resolves the
@meshsdk/core-csl / whisky-evaluator WASM path at runtime (createWallet 500).
Build the app in Dockerfile.ci and serve it with `next start` so the smoke
exercises the same production output Vercel deploys (and starts instantly
instead of compiling routes on demand). Copy prisma.config.ts before npm ci so
the postinstall generate sees the Prisma 7 config; set a dummy DATABASE_URL for
build-time module evaluation (compose overrides it at runtime).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
ci: bump node 20 -> 22 (required by mesh 2.0 + Prisma 7)
7faf23e 
mesh 2.0's whisky-evaluator (WASM) and Prisma 7's @prisma/dev/streams-local
require node >=22. On node 20 the production `next build` fails resolving the
whisky-evaluator WASM export (and streams-local warns EBADENGINE). node 22
matches what Vercel and local builds use. Bumps the app Dockerfiles and the
node-20 workflows; the node-18 scheduled jobs are left for a separate pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
ci(smoke): use node:22-slim (glibc) instead of alpine (musl)
dd5da4a 
The production `next build` fails under node:22-alpine resolving the
@meshsdk/core-csl / whisky-evaluator WebAssembly export
(`does not provide an export named 'js_evaluate_tx_scripts'`) — a musl/WASM
incompatibility. The same build succeeds on glibc (local macOS + Vercel).
Switch the CI Dockerfiles to node:22-slim (Debian) and apt for the postgres
client.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
fix(build): externalize whisky WASM packages instead of transpiling
8300f21 
webpack can't statically resolve whisky-evaluator's WASM-backed ESM named
export during `next build` on Linux (passes on macOS), so transpiling it broke
the Docker/CI build. Move whisky-evaluator + @sidan-lab/whisky-js-nodejs to
serverExternalPackages so Node `require`s their cjs/ build at runtime (loads the
WASM synchronously) and webpack never analyzes the ESM. Keep @meshsdk/core-csl
transpiled.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
fix(ci): enable --experimental-wasm-modules for the smoke build/runtime
1183cca 
The production build's page-data worker (and `next start`) load @meshsdk/core-csl
as a native ESM external, which `import`s whisky-evaluator's WASM-backed exports.
node's ESM loader needs --experimental-wasm-modules to resolve them, otherwise:
`whisky-evaluator does not provide an export named 'js_evaluate_tx_scripts'`.
Set via NODE_OPTIONS so it covers both `next build` and the runtime `next start`.
Validated: the Dockerfile.ci image now builds end-to-end on linux.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
ci(smoke): split Dockerfile so ci-runner skips next build
ad03b9e 
The app build (with the WASM-aware production next build) succeeds and the app
starts healthy — but the bootstrap step rebuilds the ci-runner service from the
same Dockerfile, re-running next build (which ci-runner doesn't need; it only
runs tsx scripts) and failing. Split Dockerfile.ci into a base stage (deps +
source) and an app stage (base + build); ci-runner targets base so the
bootstrap runs the real runtime createWallet test against the healthy app.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@QSchlegel @claude
fix(ci): set --experimental-wasm-modules on the base stage
7c3e2a2 
The ci-runner runs tsx scripts that import the Mesh SDK (core-csl -> whisky
WASM), so it needs --experimental-wasm-modules too — not just the app build.
Move NODE_OPTIONS to the shared base stage. Validated in a Linux container:
`@meshsdk/core-csl` now imports cleanly (was: "does not provide an export named
'js_evaluate_tx_scripts'").

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@QSchlegel