Firefox, |wait for visible| and exceptions

Apologies for the wall of text, there are some nuances to this issue that I hope to get across 🙂

Lately, we’ve been switching several projects from Chrome to Firefox because of unsolved issues in chromedriver. After switching, the test scripts often need some adjustments before they work again. Most of those adjustments are either small and easy or project-specific, but there is one trend that might qualify for a HSAC PR.

Namely, there are two exceptions that plague the use of |wait for visible| when running tests on Firefox: StaleContextException in general and a WebDriverException with message “TypeError: can’t access dead object”. Both have a tendency to randomly occur when waiting for something to become visible while the page content changes. This occurs frequently when waiting for e.g. a login attempt to be successful or a payment to complete.

There is a workaround for StaleContextException: if you |set search context to| any search context (even css=body will do), BrowserTest’s invoke method swallows the exception and tries again: https://github.com/fhoeben/hsac-fitnesse-fixtures/blob/2f59a14403bacce18deea99834abe574bde0dd68/src/main/java/nl/hsac/fitnesse/fixture/slim/web/BrowserTest.java#L90-L98

However, the only workaround for the WebDriverException with message “TypeError: can’t access dead object”, is to somehow make sure that the browser will not navigate while the |wait for visible| is waiting. That is not always possible (e.g. a page that refreshes until a job has finished running), so I decided to look if I could make BrowserTest swallow this exception as well. I first tried swallowing it in waitForVisible itself, but then it would still occur in TryAllFramesConditionDecorator.invokeInFrames (because of wrapConditionForFramesIfNeeded). Then I tried to swallow it at the same spot where StaleContextException is swallowed. That did the trick, and works by adding something like this to BrowserTest.invoke:

} catch (WebDriverException e) {
    if (e.getMessage().toLowerCase().contains("can't access dead object")) {
        // This is a dead object exception, just try again
        return invoke(interaction, method, arguments);
    } else {
        throw e;
    }
}

Note that this particular WebDriverException is Firefox-specific, and occurs when trying to access a DOM object when its parent document has been destroyed (i.e. navigated away from). More information at https://blog.mozilla.org/addons/2012/09/12/what-does-cant-access-dead-object-mean/.

So my question is: would you accept a PR that makes BrowserTest swallow both exceptions by default? Or at least the WebDriverException? For the StaleContextException, the workaround results in boilerplate code and makes using HSAC with Firefox a bit of a hassle, but at least it works. For the WebDriverException however, there are cases where it is necessary to be able to swallow it, and I can’t think of any use for stopping the test on it. Even if you’d want to use it to detect unwanted navigation, the low chance of it popping up (about 1 in 5 times) makes it very unreliable for that purpose. I’d argue that swallowing this one by default would be a good idea, at least when browsing with Firefox.

[JessefSpecialisterren]

Closing now that #233 has been merged 🙂

[JessefSpecialisterren]

Managed to create a test that reliably reproduces the ‘can’t access dead object’ exception and used it to create and verify PR #233!