Skip to content

Auth credential-deny e2e conversions in #1370 are unverified, and e2e is skipped for the whole prisma-8 branch #1372

Description

@borisno2

Raised during PR #1370 (issue #1255, part of #1127). Recorded because the work was merged on reasoning, not evidence, and that should be visible rather than buried.

What is unverified

PR #1370 converted seven findMany calls to .all() / .where(…).all() in packages/auth/tests/credential-field-read-deny-e2e.test.ts, as part of removing the Prisma 7 read delegates.

That suite did not run, and cannot run on the prisma-8 branch:

  1. prerequisitesPresent requires examples/starter-auth/node_modules/@prisma/adapter-better-sqlite3, which does not exist after a clean install — the examples are still SQLite-era and their conversion is Spec: Examples, scaffolder, docs and agent guidance on Prisma 8 #1129's work.
  2. CI cannot cover it either: the e2e job is gated if: github.base_ref == 'main', and every PR in this build targets prisma-8. gh pr checks shows e2e skipping throughout.

So the package's "505 auth tests passed" does not include this file. A reviewer read the conversions and judged them semantically correct — that is reasoning, not evidence, and an unrun test file is not coverage.

The suite matters more than most: it pins that a credential field's read denial holds end to end.

Why it was merged anyway

Blocking the spec on a suite that cannot execute anywhere would have stalled it indefinitely, on a dependency (#1129) that is a separate build spec. The trade was taken knowingly.

Acceptance criteria

  • credential-field-read-deny-e2e.test.ts actually executes once the examples carry a working adapter (Spec: Examples, scaffolder, docs and agent guidance on Prisma 8 #1129)
  • All seven converted call sites pass, and any semantic drift introduced by the conversion is fixed — check ordering, scoping, and empty-vs-null differences between findMany and .all()
  • The e2e job's base_ref == 'main' gate is reconsidered, so a long-lived integration branch is not silently exempt from e2e for its whole life

Notes

The third criterion is the one with reach beyond this file. An integration branch that accumulates an entire nine-spec build while every e2e run is skipped means the first real e2e signal arrives at the prisma-8 → main merge, which is the worst moment to discover a problem.


Generated by Claude Code

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    ready-for-agentFully specified, ready for an AFK agent

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions