Six edge cases, and what is still not tested
Order changes and one optional plug-in are handled; Windows, a real editor and the Docker image are not tested.
Six edge cases; behaviour stayed the same in each
Converting a real app found six places where the old hand order or a hand rule mattered. Three are about order: the old lists ran things in the order a person wrote them, and the generated list runs them in dependency order. Two are about what the build can see. One happens with or without this change.
Each card is one edge case and what was done about it:
- The Strava plug-in stays in Server.csIt is added only when a setting is on, and a module's registration cannot read settings. Unchanged behaviour (reasoned).
- Background jobs start in module orderDependencies first, not the old hand order. They are independent loops; tests pass (reasoned).
- The two stake bindings run in a new orderFriend stake before penalty. The check service runs each one that applies, separately; tests pass (reasoned).
- The check covers only our own typesFramework services (settings, logging, the DbContext) still come from Server.cs (from code).
- The Docker restore step misses the generatorLike the Domain project today; the later full build restores again. The image build is not tested (reasoned).
- Local runs rewrite the config schema fileWith or without this change, when the generate-config setting is on (measured).
None of them changed what the app does Reasoned. The tests and the start-up check cover the three order changes, because the jobs and bindings are independent of each other. The Docker step is the one to watch when the converted app is built for real.
Windows, a real editor and the Docker image are still open
These were not tested, and why:
- Windows, in any language: it needs spend
- A real editor window (VS Code, Rider): the 5-minute click list from the earlier test still applies
- Building the converted app's Docker images, and the real CI run
- A git merge driver for the snapshot, and other migration tools
- The 9 email-template unit tests: they need an npm install first, the same on both copies
Nothing was pushed: both copies of the app are local, and the spike cost 0 US dollars Measured.
Windows is the one open point that needs spend, so it has its own question. Decide thisQ5Do we test Windows too?