Testing and Development¶

Use the focused runner during development. With no group it runs three smoke checks; it does not default to the full suite:

uv run --locked python -B scripts/test.py
uv run --locked python -B scripts/test.py smoke ui
uv run --locked python -B scripts/test.py --list
uv run --locked python -B scripts/test.py ui --dry-run

Groups select existing modules or complete test classes; files with mixed responsibilities can be split between groups without moving code or breaking fixture imports. Some UI/model integration tests perform small example solves to check displayed results or editing behaviour. The dedicated engineering and surface benchmark suites are excluded from ui.

Combine groups or target one module, class or method; overlapping selections run each test once:

uv run --locked --extra mcp python -B scripts/test.py mcp surfaces
uv run --locked python -B scripts/test.py test_qt_materials.MaterialEditorTests
uv run --locked python -B scripts/test.py test_qt_theme.ThemeTests.test_multiple_editors_sync_theme_actions
uv run --locked python -B scripts/test.py --check

tests/groups.toml assigns every test class to exactly one primary group. New classes or modules without ownership, unknown selectors and duplicate primary ownership fail validation. The smoke profile selects a few methods across groups. Listing, dry runs and coverage checks use only Python's standard library and do not import unrelated tests, Qt, PyNite or optional SDKs. --collect-only imports only the selected Python tests and lists their IDs without executing them. -v prints individual results; --failfast stops after a failure. Failures and missing viewport tools return a nonzero status. This runner installs nothing.

The viewport groups invoke the existing offline scripts, using the separately installed Node/Playwright/Chromium tools described below and the same PLAYWRIGHT_MODULE / PLAYWRIGHT_BROWSERS_PATH environment settings. The docs group runs documentation unit checks; the strict site build remains a separate command so normal tests do not require Zensical. mcp needs the optional extra to include wire tests; otherwise those tests retain their existing skip behaviour.

Run the complete Python suite for changes to shared infrastructure/model/solver contracts and before releases. A passing focused group is not permission to ignore a known failure elsewhere. all includes every Python group, including smoke cases through their primary classes, but does not start browser checks, performance benchmarks or a documentation build automatically. An explicit all run compares actual test IDs with ordinary unittest discovery before execution, including inherited or dynamically supplied cases:

uv run --locked --extra mcp python -B scripts/test.py all
# Original unittest discovery remains supported:
uv run --locked --extra mcp python -B -m unittest discover -s tests

The Qt interaction tests run offscreen. The suite covers project persistence, undo/redo, result invalidation, analytical beam checks, portal-frame diagrams, inclined/reversed members, point-force/moment jumps, explicit member connections, load preservation during splitting, rejection of invalid model topology, mixed-material/section stiffness, legacy-file migration, and material/section assignment with undo.

Spatial regressions cover all six cantilever DOFs, biaxial tip moments, torsion, roll/inertia changes, oblique local/global transformations, uniform/varying loads, springs, missing out-of-plane restraints, self-weight, spatial combinations, version dispatch, units, editing history, snapshots and reports. Widget tests use PYNITEGUI_NO_WEBENGINE=1 to avoid initializing Chromium; this is a test-only switch. Visible installed-wheel checks are separate. Install the MCP wheel into a fresh environment outside the checkout, use its Python in an active desktop session, and leave QT_QPA_PLATFORM and PYNITEGUI_NO_WEBENGINE unset:

cd /tmp
/tmp/pynitegui-rc-mcp/bin/python /absolute/path/to/PyniteGUI/scripts/smoke_desktop.py --output /tmp/pynitegui-desktop-qa --hold

The script rejects editable imports and offscreen mode. It uses disposable settings/projects, QtTest clicks and application commands for generators, arrays, undo/redo, save/reopen, background analysis, governing stations, CSV/PDF, both official space frames, actual WebEngine rendering and MCP lock/unlock on an ephemeral local port. It records screenshots, dependency/renderer diagnostics and a check list in the output directory; a timeout trace helps diagnose stalls. The --hold option leaves the verified window open for human checks. File paths are supplied programmatically, so real OS file pickers, keyboard shortcuts and the interactive print dialog still require human verification. It does not certify physical printing or any platform on which it was not run. See RELEASE.md for exact observed versions and outstanding gates.

For real viewport checks, install Playwright in a separate developer environment with its Chromium browser, then run node tests/viewport3d.spec.cjs with Playwright available to Node (or set PLAYWRIGHT_MODULE to its installed package directory). The script serves local assets temporarily and closes both browser/server after testing desktop canvas pixels, framing, six signed gizmo views, keyboard reset, zoom/selection preservation, orbit, node picking, all three work-plane drawing modes, six force/moment overlays with rolled members, PNG capture and dark appearance. Screenshots go to /tmp/pynite-3d-qa by default, configurable with VIEWPORT_QA_OUTPUT. It does not launch the user's GUI. PLAYWRIGHT_EXECUTABLE_PATH optionally selects an installed Chromium-compatible test browser; otherwise Playwright uses its downloaded Chromium.

Source: GUIDE.md. This manual follows the current repository; the published rc1 wheel has a smaller feature set.