Understanding execute async script in Selenium

Question

I've been using selenium (with python bindings and through protractor mostly) for a rather long time and every time I needed to execute a javascript code, I've used execute_script() method. For example, for scrolling the page (python):

driver.execute_script("window.scrollTo(0, document.body.scrollHeight);")

Or, for infinite scrolling inside an another element (protractor):

var div = element(by.css('div.table-scroll'));
var lastRow = element(by.css('table#myid tr:last-of-type'));

browser.executeScript("return arguments[0].offsetTop;", lastRow.getWebElement()).then(function (offset) {
    browser.executeScript('arguments[0].scrollTop = arguments[1];', div.getWebElement(), offset).then(function() {
        // assertions

    });
});

Or, for getting a dictionary of all element attributes (python):

driver.execute_script('var items = {}; for (index = 0; index < arguments[0].attributes.length; ++index) { items[arguments[0].attributes[index].name] = arguments[0].attributes[index].value }; return items;', element)

---

But, WebDriver API also has execute_async_script() which I haven't personally used.

What use cases does it cover? When should I use execute_async_script() instead of the regular execute_script()?

^{The question is selenium-specific, but language-agnostic.}

Answer 1

When should I use execute_async_script() instead of the regular execute_script()?

When it comes to checking conditions on the browser side, all checks you can perform with execute_async_script can be performed with execute_script. Even if what you are checking is asynchronous. I know because once upon a time there was a bug with execute_async_script that made my tests fail if the script returned results too quickly. As far as I can tell, the bug is gone now so I've been using execute_async_script but for months beforehand, I used execute_script for tasks where execute_async_script would have been more natural. For instance, performing a check that requires loading a module with RequireJS to perform the check:

driver.execute_script("""
// Reset in case it's been used already.
window.__selenium_test_check = undefined;
require(["foo"], function (foo) {
    window.__selenium_test_check = foo.computeSomething();
});
""")

result = driver.wait(lambda driver: 
    driver.execute_script("return window.__selenium_test_check;"))

The require call is asynchronous. The problem with this though, besides leaking a variable into the global space, is that it multiplies the network requests. Each execute_script call is a network request. The wait method works by polling: it runs the test until the returned value is true. This means one network request per check that wait performs (in the code above).

When you test locally it is not a big deal. If you have to go through the network because you are having the browsers provisioned by a service like Sauce Labs (which I use, so I'm talking from experience), each network request slows down your test suite. So using execute_async_script not only allows writing a test that looks more natural (call a callback, as we normally do with asynchronous code, rather than leak into the global space) but it also helps the performance of your tests.

result = driver.execute_async_script("""
var done = arguments[0];
require(["foo"], function (foo) {
    done(foo.computeSomething());
});
""")

The way I see it now is that if a test is going to hook into asynchronous code on the browser side to get a result, I use execute_async_script. If it is going to do something for which there is no asynchronous method available, I use execute_script.