TUI verification

Historical prototype record: This page preserves the July 27, 2026 TUI experiment. It is not current supported-product guidance or a Windows-release blocker.

Local quality gates

From tui/:

uv sync --locked --extra dev
uv run pytest
uv run ruff check .
uv run mypy src
uv build --clear

Verify generated inventory drift from the repository root:

node tui/tools/generate-parity-contract.mjs --check

The contract generator must parse exactly 201 rows and 17 sections, reject unknown overrides and duplicate IDs, and leave every unmapped row as not_yet_available.

Package smoke check

Install the built wheel into a new environment rather than reusing the source environment:

uv venv /tmp/desktop-material-tui-wheel-smoke --python 3.12
uv pip install \
  --python /tmp/desktop-material-tui-wheel-smoke/bin/python \
  tui/dist/desktop_material_tui-0.1.0-py3-none-any.whl
/tmp/desktop-material-tui-wheel-smoke/bin/desktop-material-tui --version

Inspect the archive for styles.tcss, py.typed, metadata, and all three console entry points: github, dmt, and desktop-material-tui. Delete the controlled temporary environment after the check.

Automated interaction

Textual pilot tests cover focus, button presses, text entry, dialogs, panes, localization, and resizing without a display. PTY tests cover real terminal escape sequences and lifecycle. They complement but do not replace an actual mouse-reporting terminal capture.

Headless Linux acceptance

The original dated run manifest defines the full publish-mode Lowlevel MCP/Xvfb exercise. The later path-browser and Git-wrapper manifest adds packaged mouse, paste, browser, native Git, one-line install, and cleanup acceptance. Together they require:

start from a deterministic temporary Git fixture;
inspect a screenshot before sending input;
click tabs, buttons, rows, Inputs, and TextAreas;
type and edit a commit summary/body without accidentally submitting;
prove Tab/Shift+Tab/Enter/Space behavior;
scroll and resize through wide, compact, narrow, and bilingual states;
exercise literal/fuzzy/RE2 search, builder synchronization, invalid syntax, captures, Unicode, multiline, and zero-width matching;
trigger a non-blocking notice and review it in notification history;
capture only the real built TUI;
stop Xvfb/processes and remove all controlled fixtures recorded in the cleanup ledger.

Each checked interaction must cite the exact package/commit, command, terminal, viewport, and screenshot. A planned checkbox is not evidence.

Release gate

The Linux TUI is releasable only when:

the locked Python matrix passes;
Ruff, mypy, tests, parity drift, build, and fresh-wheel smoke are green;
the real Linux interaction record is complete with cleanup;
the source commit is pushed and the remote CI result is recorded;
any published package is uniquely versioned and immutable;
docs and the parity contract describe remaining gaps without calling them complete.

Python and historical-target compatibility

Git may emit strict ISO timestamps with either a numeric UTC offset or a terminal uppercase Z. Python 3.10's datetime.fromisoformat does not accept the latter, so every Git/profile-history parser normalizes only that terminal designator to +00:00. Tests must cover Z, non-zero numeric offsets, and malformed input on the oldest supported Python version.

Platform-only lock modules stay behind runtime boundaries. In particular, Linux mypy must not statically resolve the Windows-only msvcrt.locking members, and Windows checks must still exercise the same lock path.

The release workflow can be loaded from a newer default branch while a workflow_run event targets an older commit. If prepare has already marked that upstream CI target as non-publishable, TUI packaging is skipped rather than assuming the historical commit contains tui/. A publishable current target still requires the locked environment, full matrix, wheel, source distribution, and fresh-install smoke test.