There are 2 main problems with AePageObjects making it prone to flakiness,
The page polling in AePageObjects uses Capybara.using_wait_time(0) preventing Capybara from retrying errors and reloading nodes. The page polling also does not catch any errors itself causing exceptions to be potentially raised prior to waiting the asked for time. The page polling is used browser.find_document, window.change_to, and absent? / present? / and friends on element proxy.
The ensure_loaded! is difficult to write / reason about as it is sometimes called inside of page polling block (see above) and sometimes it is not. This means that implementors have to know about all the exceptions that should be caught and using AePageObjects#wait_until is not good because it can blocks for longer than what is intended by page polling.
This PR improves AePageObjects by,
Removing PagePolling and instead solely relying on the improved #wait_until. It now rescues and retries all relevant selenium / capybara / ae_page_objects exceptions. It only throws 'WaitTimeoutError' (which includes the original cause) making it easy for user to rescue errors. It lowers Capybara's default time allowing capybara to retry / reload, but also allowing errors to be handled at the higher level (and thus another chance to success). It allows itself to be called multiple times which the top-level invocation being in charge of the timeout.
The ensure_loaded! now calls is_loaded? which is meant to quickly return an answer. Subclasses should override is_loaded? now, but no waiting / rescuing is needed.
Removed #all / #find from being delegated to Capybara node. These return Capybara nodes (and in case of all not reloadable once) and it is better to be explicit about their use. Also avoids confusion with #find provided by AePageObjects::Colleciton.
There are 2 main problems with AePageObjects making it prone to flakiness,
Capybara.using_wait_time(0)preventing Capybara from retrying errors and reloading nodes. The page polling also does not catch any errors itself causing exceptions to be potentially raised prior to waiting the asked for time. The page polling is usedbrowser.find_document,window.change_to, andabsent?/present?/ and friends on element proxy.ensure_loaded!is difficult to write / reason about as it is sometimes called inside of page polling block (see above) and sometimes it is not. This means that implementors have to know about all the exceptions that should be caught and usingAePageObjects#wait_untilis not good because it can blocks for longer than what is intended by page polling.This PR improves AePageObjects by,
#wait_until. It now rescues and retries all relevant selenium / capybara / ae_page_objects exceptions. It only throws 'WaitTimeoutError' (which includes the original cause) making it easy for user to rescue errors. It lowers Capybara's default time allowing capybara to retry / reload, but also allowing errors to be handled at the higher level (and thus another chance to success). It allows itself to be called multiple times which the top-level invocation being in charge of the timeout.ensure_loaded!now callsis_loaded?which is meant to quickly return an answer. Subclasses should overrideis_loaded?now, but no waiting / rescuing is needed.#all/#findfrom being delegated to Capybara node. These return Capybara nodes (and in case of all not reloadable once) and it is better to be explicit about their use. Also avoids confusion with#findprovided byAePageObjects::Colleciton.