Bug 323692

320518@main correctly split "no such element" from "stale element reference", fixing almost 100 subtests, but the stale reference map is wiped on page navigation/refresh, as this information is stored by the AutomationSessionProxy JS instance living on the page's global object. Expected: stale element reference Actual: no such element — "Node with identifier 'node-…' was not found" Per the spec, the stale references are associated to a given navigable, surviving navigations on that navigable: > A WebDriver session has a navigable seen nodes map which is a weak map between a navigable and a set. Tests affected: imported/w3c/webdriver/tests/classic/back/back.py::test_seen_nodes[http] imported/w3c/webdriver/tests/classic/forward/forward.py::test_basic imported/w3c/webdriver/tests/classic/forward/forward.py::test_seen_nodes[http] imported/w3c/webdriver/tests/classic/navigate_to/navigate.py::test_seen_nodes[http] imported/w3c/webdriver/tests/classic/refresh/refresh.py::test_history_pushstate imported/w3c/webdriver/tests/classic/refresh/refresh.py::test_no_browsing_context imported/w3c/webdriver/tests/classic/refresh/refresh.py::test_seen_nodes[http] imported/w3c/webdriver/tests/classic/refresh/refresh.py::test_seen_nodes[https] imported/w3c/webdriver/tests/classic/refresh/refresh.py::test_seen_nodes[https coop] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_accept[alert] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_accept[beforeunload] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_accept[confirm] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_accept[prompt] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_accept_and_notify[beforeunload-None] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_default[beforeunload-None] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_dismiss[alert] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_dismiss[beforeunload] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_dismiss[confirm] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_dismiss[prompt] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_dismiss_and_notify[beforeunload-None] imported/w3c/webdriver/tests/classic/refresh/user_prompts.py::test_ignore[beforeunload] imported/selenium/py/test/selenium/webdriver/common/stale_reference_tests.py::test_old_page imported/selenium/py/test/selenium/webdriver/common/stale_reference_tests.py::test_should_not_crash_when_calling_get_size_on_an_obsolete_element imported/selenium/py/test/selenium/webdriver/common/stale_reference_tests.py::test_should_not_crash_when_querying_the_attribute_of_astale_element I have a WIP patch from a previous attempt at bug230612 (https://github.com/WebKit/WebKit/pull/61870) that keeps the stale reference map in the C++ side, keyed by frame. It could be adjusted to be a follow-up to 320518@main (e.g. using the new StaleNode error, fix conflicts, etc.).