Create a `ChromoteSession$nagivate()` method?

[gadenbuie]

trafficstars

As described in the README under Loading a page reliably, one cannot use the following pattern to reliably wait for the page to load.

# Not reliable
b$Page$navigate("https://www.r-project.org/")
b$Page$loadEventFired()  # Block until page has loaded
b$screenshot("browser.png")

Instead, we suggest an synchronous or asynchronous approach that involves correctly threading the $Page$loadEventFired() and $Page$navigate() commands.

In practice, however, most people just sys.sleep() for some reasonable or acceptable amount of time (#176).

Considering that we have a few convenience methods already, e.g. ChromoteSession$screenshot(), I'd like to propose that we offer a similar convenience method, ChromoteSession$navigate(), that automatically waits for the page load event and awaits the value (wait_ = TRUE) or returns the promise (wait_ = FALSE).

In the base case, that would simplify the above example code to just these two lines:

b$navigate("https://www.r-project.org/")
b$screenshot("browser.png")

[jonthegeek]

I think this would be very helpful! FWIW, if the page is making API calls, $Page$loadEventFired() might not be enough for the page to really be "done". I don't have it fully loaded in my memory right now, but I remember that being the source of a lot of the complexity in {apisniffer}. I never felt like what I was doing was absolutely correct, but I was using callbacks for session$Network$requestWillBeSent and session$Network$responseReceived to make sure all of the calls that fired off had received a response (or a maximum wait time was reached) before considering the page fully loaded. That might be beyond the scope of this ticket, but I wanted to bring it up (because if you CAN handle that piece, too, that would be very cool 😊 )!

[gadenbuie]

Great point @jonthegeek. I like what pupeteer does in taking a predefined lifecycle event. In R would be a wait_until argument that takes one of the following values

wait_until	Event
"load"	Waits for the load event, i.e. b$Page$loadEventFired().
"DOMContentLoaded"	Waits for the DOMContentLoaded event.
"networkIdle0"	Waits until there are no network connections for at least 500ms.
"networkIdle2"	Waits until there are no more than 2 network connections for at least 500ms.

I also really like the page.waitFor*() methods (listed here) and would love to see some of those make it into chromote. In particular

• waitForSelector() – waits for a CSS selector to have matching elements
• waitForFunction() – waits for a function to return a truthy value
• waitForNetworkIdle() – see above
• waitForNavigation() – wait for the page to navigate to a new URL
• waitForRequest() – wait for a request to a URL (or request identified by a predicate function)
• waitForResponse() – wait for a response from a URL (or a response identified by a predicate function)

[jonthegeek]

Oooh, I like waitForNetworkIdle()! I think for my specific use case I also want to wait for responses to sent events (with a timeout), but I think mixing in network idle would help deal with part of my "but is that everything?" problem.

[jonthegeek]

Oh, ha, I just saw waitForResponse() is in there, too!

[gadenbuie]

Good to know! I hadn't included waitForResponse() in my smaller list because I wasn't sure if that would be immediately useful to chromote users. Glad to get that feedback right away. (I'll edit the list above.) I do think waitForRequest() and waitForResponse() are a package deal.

[jonthegeek]

Yeah, definitely! I'd have to understand how it works a little better; what I do right now is log responses until a certain time has passed (simulating waitForNetworkIdle(), kinda), and then check that there are corresponding response events for each of those requests. I think what I'd want to do is fire off a waitForResponse() (with a timeout) as a callback on session$Network$requestWillBeSent(), disabling that callback with waitForNetworkIdle(). That assumes waitForResponse() ties (or at least can tie) to a particular request.

I very much doubt that my use case will fall neatly onto the standard path, but having the helpers would definitely make my code feel less hacky!