mirror of
https://github.com/ajisaacs/OpenNest.git
synced 2026-10-10 23:34:14 -04:00
aj
91aef95b1b
Fix DuplicateRows_CreateNestsButtonImportsOneDrawingWithCombinedQuantity timeout from Windows run 37526243901. Native trace on DesktopPC showed the sole OK button uses control ID 2, not IDOK 1. Preserve real dialog text and nest assertions; find the sole button and stop polling only after successful posting. All 392 WinForms and 49 FrontEnd tests pass in an interactive Windows session.
86 lines
4.8 KiB
Markdown
86 lines
4.8 KiB
Markdown
# Automated Windows acceptance
|
|
|
|
The `windows-desktop` job in `.github/workflows/ci.yml` runs on GitHub-hosted
|
|
`windows-2022` for every pull request and `master` push. It executes the full
|
|
WinForms and FrontEnd test projects in Release, including shown-form STA tests
|
|
and Windows child-process lifetime tests. No production service, customer
|
|
fixture, physical GPU, or CNC controller is required.
|
|
|
|
Each test host has a five-minute hang watchdog with a mini dump; the job has a
|
|
20-minute limit. A failure in either suite fails the step, without suppressing
|
|
the other suite. This job is separate from the existing Linux `tests` aggregate;
|
|
require `windows-desktop` as well when assessing acceptance.
|
|
|
|
## Results and coverage contract
|
|
|
|
Download the `windows-desktop-test-results` artifact from the workflow run.
|
|
It contains both TRX files, any hang diagnostics, and:
|
|
|
|
- `windows-acceptance.json`: commit/tree, runner platform, workflow identifiers,
|
|
TRX hashes, enumerated results, failures/skips and per-area test evidence.
|
|
- `windows-acceptance.md`: a readable summary, also shown in the Actions job
|
|
summary, with remaining acceptance beside the automated result.
|
|
|
|
Artifacts are retained for 14 days. Save evidence needed for longer-lived task
|
|
verification before expiration. On pull requests the checked-out commit can be
|
|
GitHub's test merge commit; use the report's commit/tree and workflow link, not
|
|
just the branch name.
|
|
|
|
`scripts/windows-acceptance.json` is the committed coverage contract. It names
|
|
required test methods and the minimum number of executed cases for each
|
|
(including theory rows). The report fails if either TRX is absent/malformed,
|
|
empty or inconsistent; if a required method/case is missing, failed or skipped;
|
|
if any other test fails; or if a run was aborted. Skips outside the named contract
|
|
are retained explicitly, never counted as passes. Added cases are allowed;
|
|
removing or renaming required coverage requires an intentional manifest change.
|
|
The whole suites still run, not just this manifest's subset.
|
|
|
|
The areas are BOM import, Nest Info material preservation, saved-nest browsing,
|
|
nesting/commit policy, PDF export, the cutting dialog, automatic cutoffs,
|
|
overlap/posting consent, UI lifetime, import/export failures, and MCP processes.
|
|
The manifest is public test coverage only. Private tracker IDs, plans, customer
|
|
data, and acceptance notes stay out of the repository and CI artifacts.
|
|
|
|
The report is evidence, not automatic task closure. A passing area does not
|
|
establish absent implementation, real-service/two-PC behavior, DPI or physical
|
|
printing quality, packaged-build behavior, or machine safety. In particular,
|
|
fake-repository browser tests do not verify the HTTP/SQLite save/paging lifecycle.
|
|
Keep those remaining gates in the tracker; link exact test names and the workflow
|
|
run when updating an individual task.
|
|
|
|
## Repeat locally on Windows
|
|
|
|
Use a clean, committed checkout, .NET 8 and Python 3. Run from the repository root
|
|
in PowerShell in a logged-in Windows desktop session: an OpenSSH/service session
|
|
(session 0) cannot show the real native completion dialogs these tests exercise.
|
|
A remote interactive-session launch should use an explicitly approved temporary
|
|
scheduled task, without saved passwords, and remove that task after completion.
|
|
Test windows may appear briefly; do not manually acknowledge them during a run.
|
|
Always use a new results directory so stale TRX files cannot stand in for a failed
|
|
build. Do not reuse downloaded CI results as a new local run.
|
|
|
|
```powershell
|
|
$results = Join-Path 'TestResults' ([guid]::NewGuid().ToString())
|
|
$failed = @()
|
|
foreach ($project in @('OpenNest.WinForms.Tests', 'OpenNest.FrontEnd.Tests')) {
|
|
dotnet test "$project/$project.csproj" -c Release --blame-hang-timeout 5m --blame-hang-dump-type mini --logger "trx;LogFileName=$project.trx" --results-directory $results
|
|
if ($LASTEXITCODE -ne 0) { $failed += $project }
|
|
}
|
|
python scripts/windows_acceptance.py --results $results
|
|
if ($LASTEXITCODE -ne 0 -or $failed.Count -gt 0) { throw 'Windows acceptance failed; inspect results.' }
|
|
```
|
|
|
|
The CLI refuses Linux/macOS as Windows runtime evidence and refuses tracked
|
|
source edits. The source revision is provenance of the checkout running the
|
|
command, not a cryptographic attestation of arbitrary supplied TRX files; the
|
|
fresh-checkout CI run and the fresh-directory local procedure establish that
|
|
relationship. The JSON retains hashes so the saved files can be matched later.
|
|
|
|
## Maintain the gate
|
|
|
|
Run `python scripts/test_windows_acceptance.py -v` on any platform after changing
|
|
the parser or manifest. Its generated TRX fixtures test the reporter only, not
|
|
the application. Exercise a draft PR on Windows before integrating new form
|
|
cases. Inspect actual TRX failures and skip reasons; a Linux cross-build is not
|
|
a Windows pass. Keep visual/operator limitations alongside the named cases.
|