Skip to content

test(ci): land Windows-first packaged desktop E2E and registered browser quarantine #133

Description

@slashdevcorpse

Context

PR #130 prototyped cross-platform unit/browser lanes, a quarantine registry, and packaged Electron E2E, but its exact head 84cdefe5fbbe49ff7bb4aebea3259da53855bd5b is stale-base and not merge-qualified. It is intentionally deferred from the Windows .9 release.

Hosted evidence on that head:

  • required quality failed;
  • e2e_windows failed 2 of 7 journeys (Codex provider missing during readiness; terminal search remained No results);
  • e2e_linux failed all 7 journeys;
  • unit_windows failed 3 Windows Job smoke tests;
  • unit_macos failed 15 tests;
  • codecov/patch failed;
  • five review threads remain unresolved, including workflow-contract allowlist bypasses.

Scope

Rebuild the useful test infrastructure on a fresh branch from current main, Windows first:

  • run the complete unit suite in the existing Windows quality model;
  • make the stable browser suite blocking on Windows;
  • keep quarantine entries machine-readable, explicitly registered, bounded, and non-blocking only for the exact registered command;
  • run offline packaged Windows Electron journeys with isolated homes/ports/workspaces, failure-only redacted diagnostics, and identity-verified test-process cleanup;
  • preserve existing required Linux/macOS coverage without expanding macOS work in this goal.

This issue is test/CI only. Do not carry product runtime changes from PR #130.

Existing issue routing

Acceptance

  • Fresh-main PR contains only test, fixture, package, and workflow changes approved for this goal.
  • The Windows packaged E2E lane passes all seven journeys on the immutable hosted head and passes a repeat run without assertion weakening.
  • Stable Windows browser tests block; only exact registered quarantine commands may continue on error.
  • Workflow validation binds matrix-runner exceptions to the intended job and rejects appended failure-masked commands.
  • E2E network guards tolerate malformed completion URLs, diagnostics redact secrets/paths/content, and cleanup reports unknown/failure rather than false success.
  • Quarantine validation has one non-duplicated source of truth and rejects future dates.
  • All required exact-head checks pass with zero unresolved review threads.
  • No release publishing, installer qualification, or product bug fix enters the diff.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions