stablyai--orca/.github/workflows/win-update-e2e.yml

name: Windows Update-Survival E2E

# Why: proves what happens to terminal sessions across a real Windows app update
# by installing one released version, updating to another, and asserting on the
# packaged artifacts (daemon survival, terminal interactivity, zero console
# flashes). This MUST run on a disposable CI Windows: electron-builder's oneClick
# installer uninstalls the registry-registered copy of the app before every
# install, so running the harness on a machine with a real Orca install would
# delete it. A fresh runner has no Orca registered, so it is the only safe home
# for install-based testing. Manual dispatch only.

# Why: workflow_dispatch only works once a workflow is on the default branch.
# To exercise this from the unmerged feature branch WITHOUT touching main, a
# push trigger scoped to that exact branch runs the workflow from the branch's
# own tree. The paths filter keeps it to harness/workflow edits so ordinary
# commits don't spend 30 min of Windows runner time. Push runs have no inputs,
# so every parameter falls back to a default below.
on:
  push:
    branches:
      - Jinwoo-H/windows-update-survival
    paths:
      - 'tools/win-update-e2e/**'
      - '.github/workflows/win-update-e2e.yml'
  workflow_dispatch:
    inputs:
      from_tag:
        description: Release tag to install first (e.g. v1.4.124-rc.8)
        required: true
        type: string
      to_tag:
        description: Release tag to update to (e.g. v1.4.124-rc.9)
        required: true
        type: string
      expect:
        description: Expected outcome profile
        required: true
        type: choice
        default: cold-restore
        options:
          - cold-restore
          - survival
      asset_pattern:
        description: Installer asset glob
        required: false
        type: string
        default: '*windows-setup.exe'
      soak_seconds:
        description: Post-relaunch console-window watch duration
        required: false
        type: string
        default: '60'

permissions:
  contents: read

# Why: cancel a superseded run when iterating — a new push to the branch makes
# the in-flight Windows job obsolete, so free the runner immediately.
concurrency:
  group: win-update-e2e-${{ github.ref }}
  cancel-in-progress: true

jobs:
  update-e2e:
    name: update ${{ inputs.from_tag || 'v1.4.124-rc.8' }} -> ${{ inputs.to_tag || 'v1.4.124-rc.9' }} (${{ inputs.expect || 'cold-restore' }})
    runs-on: windows-2022
    timeout-minutes: 30
    env:
      FROM_TAG: ${{ inputs.from_tag || 'v1.4.124-rc.8' }}
      TO_TAG: ${{ inputs.to_tag || 'v1.4.124-rc.9' }}
      EXPECT: ${{ inputs.expect || 'cold-restore' }}
      ASSET_PATTERN: ${{ inputs.asset_pattern || '*windows-setup.exe' }}
      SOAK_SECONDS: ${{ inputs.soak_seconds || '60' }}

    steps:
      - name: Checkout
        uses: actions/checkout@v6
        with:
          # This job only reads the repo and downloads a release; it never pushes.
          persist-credentials: false

      - name: Setup Node.js
        uses: actions/setup-node@v6
        with:
          node-version-file: package.json

      - name: Setup pnpm
        uses: pnpm/action-setup@v6
        with:
          run_install: false

      # Why: the harness only needs its runtime deps (the Playwright Electron
      # driver); it drives already-built installer artifacts, so no app build.
      - name: Install dependencies
        run: pnpm install --frozen-lockfile

      # Why: fail fast with a readable message if a tag has no matching Windows
      # installer, rather than deep inside the harness.
      - name: Download installers
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        shell: pwsh
        run: |
          New-Item -ItemType Directory -Force artifacts/from | Out-Null
          New-Item -ItemType Directory -Force artifacts/to | Out-Null
          gh release download "$env:FROM_TAG" --repo "${{ github.repository }}" --pattern "$env:ASSET_PATTERN" --dir artifacts/from
          gh release download "$env:TO_TAG" --repo "${{ github.repository }}" --pattern "$env:ASSET_PATTERN" --dir artifacts/to
          $from = Get-ChildItem artifacts/from -Filter *.exe | Select-Object -First 1
          $to = Get-ChildItem artifacts/to -Filter *.exe | Select-Object -First 1
          if (-not $from) { throw "No installer matching '$env:ASSET_PATTERN' in $env:FROM_TAG" }
          if (-not $to) { throw "No installer matching '$env:ASSET_PATTERN' in $env:TO_TAG" }
          "FROM_EXE=$($from.FullName)" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append
          "TO_EXE=$($to.FullName)" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append

      # Why: no --install-dir here. A CI runner has no real Orca registered, so
      # the harness installs to the default per-user location and owns it — the
      # isolated-install machinery exists only to (imperfectly) protect a real
      # developer install, which does not apply on a throwaway runner.
      - name: Run update-survival harness
        id: harness
        shell: pwsh
        env:
          # Why: on a driving failure the harness dumps a screenshot + DOM state
          # here so the actual post-onboarding UI is visible in the artifacts.
          ORCA_E2E_DIAG_DIR: artifacts/diag
        run: |
          $log = "artifacts/harness-output.log"
          node tools/win-update-e2e/run.mjs `
            --from "$env:FROM_EXE" `
            --to "$env:TO_EXE" `
            --expect "$env:EXPECT" `
            --soak-seconds "$env:SOAK_SECONDS" 2>&1 | Tee-Object -FilePath $log
          exit $LASTEXITCODE

      - name: Upload harness output
        if: always()
        uses: actions/upload-artifact@v7
        with:
          name: win-update-e2e-output
          # Why: log + diagnostics only, never the multi-hundred-MB installers.
          path: |
            artifacts/harness-output.log
            artifacts/diag/**
          retention-days: 7
          if-no-files-found: warn