Interact with cross-origin iframe

[timwoocker]

When loading a page which has an iframe with a different origin, page.frames() either doesn't return the frames at all or the url Option is None (idk why it behaved differently sometimes).
Either way, the only way to access a frame's content is by using GetTargetParams which successfully returns the targets of all frames. Unfortunately - since frames are not "pages", I can't use browser.get_page().
The only way I see to interact with a frame is using AttachToTargetParams which returns a SessionId. However, I don't see a way to use the SessionId other than creating command from raw json.

Is there a nice way to achieve this already?
If not - I feel like a utility method which creates a Session or Target or Frame object which shares some functionality with the Page struct would be a nice addition. Any thoughts on that?
An even simpler alternative would be something like a browser.execute_with_session() function where a SessionId can be passed.

[Sytten]

If you are willing to do a PR for that, I will look at it. Right now I dont really have time to check that.

[timwoocker]

I could work on a PR but I need some advice since there are some things I don't understand.
Here is what I do and observe in-order:

1. I load a page using browser.goto() which has a cross-origin iframe (and wait for it to load completely).
2. Trying page.frames() doesn't return the frame at all. browser.pages() also doesn't return the frame.
3. page.execute(GetTargetsParams::default()) returns the frame's TargetInfo.
4. The TargetInfo has attribute attached set to true. This is weird because apparently it's not actually attached?
5. Since the target doesn't seem to be attached, I call browser.execute(AttachToTargetParams::builder().flatten(true).target_id(t.target_id.clone()).build()?) which works and returns a SessionId.
6. I added an execute_on_session(cmd, session_id) function to Browser which simply uses CommandMessage::with_session instead of CommandMessage::new. I use this function to execute with GetDocumentParams to check if I can access the DOM. This works correctly as expected - no problems at all. But I can't use any utility functions of Page and Element - since I don't have a page but just the session id.
7. For some reason, calling browser.get_page() on the frame's target id after executing AttachToTarget returns a Page. Its session id is the same as the one returned by AttachToTarget.
8. Using the page returned by get_page (calling find_element or basically any other function) doesn't work. It times out after a few seconds.

So my guess is that calling AttachToTarget manually is the wrong approach? Or the handler doesn't correctly register the created target?
I also wasn't able to use SetAutoAttach on the outer page - it just didn't change anything.

So the only thing I was able to do successfully was adding the execute_on_session function on the browser. But I have no idea what's wrong with the page and what to do about it. Any ideas?

[timwoocker]

Some interesting observations:

1. The TargetType of the iframe's page is "iframe". Which is not an enum variant so it defaults to Unknown("iframe")
2. HandlerMessage::GetPages filters by calling is_page() which returns false for the iframe page. So the iframe page is never visible from the pages() function on Browser.
3. HandlerMessage::GetPage however doesn't check if the type is page. That's why I was able to get the page object for the iframe.
4. The poll() function of Target firstly checks if its type is "page". That's why I got the timeout.
5. Changing the code so that poll works on either "page" or "iframe" works to some degree - I can now call functions like find_element and it actually returns the element inside the iframe.
6. Trying to evaluate javascript, however, fails with the message Chrome(Error { code: -32602, message: "Either objectId or executionContextId or uniqueContextId must be specified" })
7. Also, when poll also pools iframe targets, the iframe doesn't visually load for some reason. I was still able to use find_element, so the DOM still gets loaded. It's just invisible somehow? I really don't understand, why.

[Sytten]

Honestly this is over my knowledge at this point. Maybe @mattsse knows more. If I have time I might be able to look into it. I know I will need to work with iframe sometime next year so I will probably hit a similar bug.