[bluzername]

Fixes part of #58 (bug 7, "A Command shortcut leaves Command held down").

press(key, command=True) set the Command flag on the key's own events but never send anything for the Command key itself. On macOS 27 the system then keep Command down after the press (CGEventSourceFlagsState read it held until something send a Command key-up). After that, every letter from type_text arrive as Command+key so nothing reach the document, and the next click is a Command-click. It follow go_back (Command-[) and clear_field (Command-A) too.

Fix: send Command's own key down before the key and up after it, in a finally so an interrupted press still release it.

-    _down_then_up(event)
+    if not command:
+        _down_then_up(event)
+        return
+    try:
+        _post(_command_key(True))
+        _down_then_up(event)
+    finally:
+        _post(_command_key(False))

Test (TDD): added two tests in tests/test_input_abort.py:

• test_a_command_shortcut_sends_the_command_key_down_before_and_up_after_the_key checks the full down/up sequence around a Command shortcut.
• test_the_command_key_goes_back_up_even_when_posting_it_is_interrupted checks the Command key still goes back up when posting it get interrupted.

Both fail on main (only 2 events post, the key's own, with the flag set but no separate Command key events) and pass with the fix. Renamed the old test_a_key_goes_back_up_when_interrupted_after_key_down to test_a_plain_key_goes_back_up_when_interrupted_after_key_down and dropped command=True from it, since it was really testing the plain key path, and its old assertion (one event pair with the flag set, no separate Command events) is not what command=True produce anymore with this fix.

Verification: uv run ruff check . && uv run ruff format --check . clean, uv run pytest -q: 732 passed, 2 skipped, 1 xfailed (main has 730, net +2 tests here).

🤖 Generated with Claude Code