34 lines
2.1 KiB
Markdown
34 lines
2.1 KiB
Markdown
# Owner-local test strategy
|
|
|
|
References inspected before migration:
|
|
- Current Makepad `libs/makepad_test/{README.md,src/runtime.rs,src/app_process.rs}`.
|
|
- Makepad `examples/counter/tests`: actual locator wait/click/assert APIs.
|
|
- Nigig `crates/apps/nigig-mpesa/tests/ui.rs`: direct makepad-test UI tests.
|
|
- Nigig `crates/apps/spreadsheet/spreadsheet-ui/tests/ui.rs`: per-test configuration.
|
|
- Nigig `crates/robius-sms/tests/sms_pipeline.rs`: public API integration tests and
|
|
explicit separation of host tests from device-only JNI/permission claims.
|
|
|
|
Nigig's older framework notes refer to StudioHub and ignored headless tests.
|
|
The current Makepad crate owns a release app process and communicates over its
|
|
remote protocol. Budget uses that current API, with no ignored tests and a Linux
|
|
display/Xvfb. It does not copy stale hub setup or substitute widget-count mocks.
|
|
|
|
`app/tests/ui.rs` drives history navigation, date text inputs, validation buttons
|
|
and capture-boundary messaging. Each case gets isolated HOME/XDG directories;
|
|
no real user budget is modified. `app/tests/models.rs` tests the public application
|
|
library, replacing the nested, duplicated-source model harness.
|
|
|
|
`budget-capture/tests/ocr.rs` executes the real embedded models against a generated
|
|
product image and a blank image, and checks invalid-buffer rejection. Its fixture
|
|
is synthetic, not a user screenshot. `tests/java/ProductExtractionTest.java` retains
|
|
12 price/title/currency tests, compiled by javac/JUnit directly. Contract Python
|
|
tests are explicitly static guards, not device tests.
|
|
|
|
`app/tests/run-clean.sh` creates a disposable Rust environment following Nigig's
|
|
clean-runner pattern. It runs the owner-local suites, builds/links/runs the native
|
|
app, and with FULL_ANDROID=1 installs the Makepad SDK, builds the actual APK and
|
|
inspects its classes, manifest, exported JNI function and embedded models. It
|
|
cleans up even when a command fails. Optional VALIDATION_ARTIFACTS preserves UI
|
|
failure diagnostics; VALIDATED_LOCKFILE_OUTPUT writes the resolved lockfile.
|
|
|
|
Android hardware/emulator acceptance remains a separate gate. See ANDROID_CAPTURE.md.
|