Not all `locator` APIs currently retry

[ankur22]

Brief summary

While working on #5111 and #5163, I came to realise that locator APIs aren't retrying during specific errors -- when an element is not visible or when an element is not attached to the DOM.

Like we have done for the locator APIs that require a pointer such as click (#5111 and #5163), we need to add a retry mechanism for the other APIs, specifically for the ones that work with newAction.

Take a look at retryPointerAction and how it used in newPointerAction to understand a possible solution for newAction.

The affected locator APIs are:

• Locator
• BoundingBox
• Clear
• Fill
• Focus
• GetAttribute
• InnerHTML
• InnerText
• TextContent
• InputValue
• SelectOption
• Press
• PressSequentially
• Type
• DispatchEvent
• IsChecked
• IsEditable
• IsEnabled
• IsDisabled
• Evaluate
• EvaluateHandle

k6 version

NA

OS

NA

Docker version and image (if applicable)

NA

Steps to reproduce the problem

NA

Expected behaviour

Retry when an element is not visible or connected to the DOM until the timeout is reached.

Actual behaviour

Early exist with an error when the element is not visible or connected to the DOM.

[janHildebrandt98]

Hi,

is this still relevant? If so, I would like to work on it :)
I saw that the linked PR #5595 was closed unmerged...
Thanks!

@ankur22

[ankur22]

@janHildebrandt98 please go ahead 👍 A quick note on AI usage: If you do use AI, please make sure to review your code and ensure that the behaviour matches Playwright. So you will need to create a Playwright test script which works, and a k6 browser test script that doesn't work. After the fix has been implemented, the k6 browser test script should work. Thank you!

[janHildebrandt98]

I picked up #5572 in the meantime.
Will start working on the retry issue when that's taken care of 👍

[janHildebrandt98]

@ankur22
Hi,
I worked on this quite a bit now but I have trouble understanding what the actual issue is:
As I understand it we want to avoid the situation where, by some parallel action, the element we are going to execute logic on gets hidden/detached after the actionability checks but before we run our logic. This means after:

k6/internal/js/modules/k6/browser/common/element_handle.go

Lines 1589 to 1593 in 6d8bf30

	if !force {
	if _, err := h.waitForElementState(apiCtx, states, timeout); err != nil {
	return nil, err
	}
	}

But before the call to fn(). Because otherwise h.waitForElementState() would block and we would get a timeout error which is what playwright does so what we want too, right?

This is where I am confused:
Apart from .boundingBox(), most locator APIs don't throw when the element is hidden/detached because the underlying js doesnt care. It just fails silently (focus(), fill()...) hence we don't get an error but nothing happens either. I did some testing with playwright and it seems to be the same there.

Please help me out here :)
PS: I did not test all of the APIs you listed, maybe it behaves differently in some of them?

[ankur22]

@janHildebrandt98 Sure, let's work together on this.

As I understand it we want to avoid the situation where, by some parallel action, the element we are going to execute logic on gets hidden/detached after the actionability checks but before we run our logic. This means after:

k6/internal/js/modules/k6/browser/common/element_handle.go

Lines 1589 to 1593 in 71a7ea7

	if !force {
	if _, err := h.waitForElementState(apiCtx, states, timeout); err != nil {
	return nil, err
	}
	}

But before the call to fn(). Because otherwise h.waitForElementState() would block and we would get a timeout error which is what playwright does so what we want too, right?

This sounds correct to me.

Apart from .boundingBox(), most locator APIs don't throw when the element is hidden/detached because the underlying js doesnt care.

I think I understand what you're saying. The only reason why we're seeing the detach errors in newPointerAction is because we're calling CDP APIs using the element's object ID, whereas with most of the listed APIs listed in the issue description (e.g. focus, fill, etc) we're not calling a CDP API using the element's object ID.

It just fails silently (focus(), fill()...) hence we don't get an error but nothing happens either. I did some testing with playwright and it seems to be the same there.

I understand that we might not get the same errors back that we're seeing in newPointerAction, but are we not seeing any other errors being returned? I'm kind of surprised that these function calls in the injected script are silently failing. For example the focus function in the injected script should return an error if we try to work with it when it has been hidden or detached. how did you verify that?

Thanks for this discovery and what seems like thorough research so far! 🙂