DomConverter should not trim whitespaces in nodes that neighbour with black–box elements (e.g. MathML)

[oleq]

📝 Provide detailed reproduction steps (if any)

Note: I used my PoC code here that brings necessary conversion to load <math> to showcase the problem. But the issue isn't about the PoC code because it occurs before any conversion kicks in, which is at the raw HTML parsing stage.

1. Try to load this kind of HTML into the editor

<p>Foo <math xmlns="http://www.w3.org/1998/Math/MathML">x</math> bar</p>

2. It loads just fine
   
3. But now try this one (note the trailing space in the <math>):

<p>Foo <math xmlns="http://www.w3.org/1998/Math/MathML">x </math> bar</p>

✔️ Expected result

Spaces after "foo" and before "bar" should be preserved.

❌ Actual result

The later is gone:

📃 Other details

I debugged the issue, and the story is as follows:

1. In DomConverter there's a private method that converts native DOM nodes into text.
2. It does trimming from both sides if
   • there's no adjacent node,
   • adjacent node is an element,
   • adjacent node is text and its last character is a white-space (this is is the case we're entering here)
3. The trimming is based on what is considered a "touching inline node".
4. The "touching inline node" is found using the native tree walker.
5. The walker enters a <math> and if it ends with some white spaces like, for instance, an indentation used to make the code human-readable (or like a trailing space from the TC), it figures this is it and gives a green light for trimming, which is obviously wrong.

The white spaces in the <math> do not render like regular white spaces in a bold or something, so they should never be considered.

We need a fix and here are some ideas:

• Some DomConverter#objectLike collection that would collect black–box elements. It would be used in the walker to prevent it from considering black–box content and thus, it would prevent the trimming.
• I think we could safely skip all elements in the tree walker that have Element#namespaceURI. I didn't test it but I suppose if we replaced <math> with inline <svg> (that is also formatted inside), we'd end up with the same issue.

The issue was spotted when creating a PoC for MathML .

---

If you'd like to see this fixed sooner, add a 👍 reaction to this post.

[oleq]

WDYT @scofalik?