Testing GUIs is hard: unit tests require a tremendous amount of mocking and often don't capture the exact user input anyway, integration tests on the other hand are difficult to setup as they often require a dedicated infrastructure with virtual machines.

libuitest should become something like the image recognition part of openQA: it receives a video or image stream from some source and searches for matches (e.g. button found, sound played). It should also be able to send clicks, key presses, strings, etc. back via an abstract API to a backend.

It should be compatible to openQA's needles, but the main way how to write tests should be done via a more generic API: `

help_menu = find_menu("Help", search_area_percent=((0, 0), (100, 0)))

help_menu.click()

about_my_program_button = find_menu("About Foo", search_area_percent=(help_menu.location, (100, 0)))

about_my_program_button.click()

window = find_window("About Foo")

about_foo_text = get_text_from_area(window.area)

assert(about_foo_text == f"This is Foo version {env.FOO_VERSION}")

The image recognition part should be handled by OpenCV and the ocr via tesseract. It would be also nice to use some simple machine learning so that libuitest could be trained to find buttons, menus, windows, etc. independently of the used GUI and theme (or at least robustly if you tell it which theme & GUI you are using).