[testdriver] Add an execute_script function to testdriver.

This provides a direct way to execute script in another browsing context.

Author: jgraham

rfcs/execute_script.md

@@ -0,0 +1,82 @@
+# RFC 89: `testdriver` Add an `execute_script` method.
+
+## Summary
+
+Add an `execute_script` method to testdriver that takes a js function,
+arguments to pass to the function, and the context id to execute it it
+in a remote context. If the function returns a promise we return the
+resolved value of the promise, otherwise we return the return value of
+the function.
+
+## Details
+
+Sometimes a test window wants to check the status of something in a
+remote window. It's possible to do this in a number of ways, for
+example by posting messages from the remote window to the test
+window. However in cross-origin cases messge passing on the js side
+can be difficult or impossible. In addition when tests are defined in
+the test window, it's useful to avoid putting complex logic in other
+browsing contexts because these won't have the same error handling
+properties as the test window.
+
+A simple way to solve these problems is to provide a mechansim to
+execute script in a different context. This isn't possible using
+content APIs alone, but is possible in WebDriver, and we can reuse
+that to define a `execute_script` function that works across contexts:
+
+```
+execute_script(fn, args, context)
+```
+
+`body` is a js function for the script to execute. To pass it to the
+remote context, this is converted to a string using
+`toString()`. `args` is null or an array of arguments to pass to the
+function on the remote side. Arguments are passed as JSON. `context`
+is an object that can be resolved as a testdriver context id to
+identify the remote context (see [RFC
+88](https://github.com/web-platform-tests/rfcs/pull/88). If the return
+value of the function when executed in the remote context is a promise
+the promise returned by `execute_script` resolves to the resolved
+value of that promise. Otherwise the `execute_script` promise resolves
+to the return value of the function.
+
+The wptrunner implementation uses [Execute Async
+Script](https://w3c.github.io/webdriver/#execute-async-script). We
+wrap the provided script text and arguments into an `Execute Async
+Script` call as:
+
+```
+let callback = arguments[arguments.length - 1];
+let rv = ({function_string}).apply(null, {json.dumps(args)});
+return Promise.resolve(rv).then(callback)
+``
+
+## Example
+
+```
+<iframe src="child.html"></iframe>
+<script>
+setup({single_test: true});
+onload = {
+  let value = test_driver.execute_script(async (elemId) => {
+  await new Promise(resolve => onload(resolve))
+  return document.getElementById(elemId).textContent
+}, ["test"], iframe.contentWindow);
+  assert_equals(value, "PASS");
+  done();
+}
+</script>
+```
+
+## Risks
+
+WebDriver is single-threaded, so all the WebDriver-based actions need
+to be queued. That means that waiting for a promise in a remote
+context blocks all testdriver functionality that uses WebDriver, and
+also blocks the harness from returning results. This implies a high
+risk of timeouts.
+
+## References
+
+[PR 29803](https://github.com/web-platform-tests/wpt/pull/29803)
+contains a prototype implementation of this.