fix: patchChildSlotNodes & scopedSlotTextContentFix not being applied

[johnjenkins]

What is the current behavior?

GitHub Issue Number: Fixes #6054

TLDR ... customElement.childNodes, customElement.children monkey-patches are never actually switched-on in Stencil atm. scopedSlotTextContentFix doesn't fully patch the textContent.

...

Contrary to documentation, the experimentalSlotFixes or slotChildNodesFix flags have no effect on the childNodes or children accessors. The patches are only applied if the deprecated needsShadowDomShim flag is fulfilled.

This must be a mistake / a hangover from supporting legacy browsers.

Additionally scopedSlotTextContentFix alone doesn't fully patch the host.textContent accessor. It adds a half-way patch - which I believe to be more confusing than useful.

To get the full textContent patch users much add both experimentalScopedSlotChanges AND scopedSlotTextContentFix (experimentalScopedSlotChanges alone won't add ANY patch).

What is the new behavior?

Adding experimentalSlotFixes or slotChildNodesFix now patches childNodes, children and childElementCount - they only return the nodes / elements that have been slotted - not the items within the component internals. This more closely mimics shadowRoot enabled components.

I have removed the experimentalScopedSlotChanges for the textContent patch.
Users need only apply scopedSlotTextContentFix or the unified experimentalSlotFixes.

Additionally, I have rolled-in the experimentalScopedSlotChanges flag within the unified experimentalSlotFixes flag - I believe this makes sense as experimentalScopedSlotChanges will eventually be 'on' by default.

Documentation

Does this introduce a breaking change?

• Yes
• No

Testing

• Added some spec tests for dom-extras.
• Amended wdio tests (e.g. the text-content-patch test now only tests for the slotted text and doesn't return internal text)
• I made some flaky wdio tests slightly more reliable on my local machine. Happy to remove these as out-of-scope.

Other information

Making these changes did open a slight a can of worms because the codebase has never actually dealt with host.childNodes being monkey patched; everything assumes it will return a full list of component internals...

• Within non-shadow vdom slot handling
• Within other monkey-patched methods

To fix this, using the precedent of appendChild (and similar methods) within dom-extras.ts, I have added a __childNodes accessor to return a component's 'internal' nodes.

[johnjenkins]

@christian-bromann any insight into the erroring windows unit test?
i.e. Is it genuine or just a flaky, one-off thing?
🤞for the latter as I don't have an easy way of running a comparable windows vm :D

EDIT
As per comments here < pretty sure it's my use of beforeAll (it's the only jest beforeAll in the codebase so makes sense :) - refactored)

[christian-bromann]

i.e. Is it genuine or just a flaky, one-off thing?

unfortunately yes, I feel like it may be worth transitioning to Vitest but I haven't had the time to look into that nor find out why it is throwing that error.

[christian-bromann]

Awesome stuff 👏