click_element races Bootstrap toasts, so any click within 5s of a save is a coin flip

[jaredlockhart]

Problem

Base.click_element waits for EC.element_to_be_clickable and then clicks:

def click_element(self, locator):
    el = self._wait_clickable(locator)
    el.click()

element_to_be_clickable checks that the element is visible and enabled, not that it is unobscured. The app renders Bootstrap toasts into a fixed #toasts-container after saves, initialised as new bootstrap.Toast(toastEl) with no options, so they inherit the defaults autohide: true, delay: 5000 and float over the page for five seconds. Any click landing in that window passes the wait and then throws:

selenium.common.exceptions.ElementClickInterceptedException: Element
<button class="btn p-0 border-0 text-secondary" type="button"> is not clickable
at point (1297,640) because another element <div class="toast-body"> obscures it

There is no toast handling anywhere in the integration suite: git grep -i toast experimenter/tests/integration returns nothing. click_element is the shared click path for the whole suite, so this is not specific to one test.

Observed on the Desktop Nimbus UI (Release Firefox) job taking a retry attempt on test_rollouts_ui.py::test_rollout_can_be_launched[FIREFOX_DESKTOP], raised from pages/base.py click_element via pages/rollouts/form_base.py edit. The earlier CI profiling work counted 39 of these in a single bad run.

Proposed change

In experimenter/tests/integration/nimbus/pages/base.py, retry the click when it is intercepted, and dismiss any visible toast before retrying so the retry does not have to wait out the full 5 second autohide:

• add ElementClickInterceptedException to the selenium.common.exceptions import
• add a _dismiss_toasts helper that removes the show class from .toast.show via JS, wrapped in contextlib.suppress (already imported)
• rewrite click_element to attempt the click inside a short bounded WebDriverWait, dismissing toasts and returning False on ElementClickInterceptedException

The dismissal only runs after a real interception, so it never speculatively mutates page state, and it only touches .toast. No test asserts on toasts, so nothing loses coverage. A permanently obscured element still fails once the bounded wait expires, so this does not mask a genuine regression where something legitimately covers a control.

Out of scope

There are roughly 17 raw .click() call sites in page objects that bypass click_element. Those keep the current behaviour. This change covers the shared path, which is where the failure has actually been observed.

Acceptance criteria

• An intercepted click retries rather than failing immediately, bounded so a permanent overlay still fails.
• Visible toasts are dismissed only on the retry path, never pre-emptively.
• Desktop Nimbus UI stops failing with ElementClickInterceptedException naming toast-body, confirmed by sampling job logs on main after merge.

┆Issue is synchronized with this Jira Task