Opened 3 weeks ago
#61560 new defect (bug)
HTML API: Fragment parser should not add context node to stack of open elements.
| Reported by: |
dmsnell
|
Owned by: | |
|---|---|---|---|
| Milestone: | 6.7 | Priority: | normal |
| Severity: | normal | Version: | 6.6 |
| Component: | HTML API | Keywords: | |
| Focuses: | Cc: |
Description
The HTML Processor currently only supports a fragment parser whose context node is <body>. As it starts supporting more context nodes it's important that the fragment interface works as expected, but it's currently wrong in that it adds the context node to the stack of open elements when it shouldn't.
The visible impact of this is that breadcrumbs currently report HTML > BODY > …. It's likely that instead it should skip BODY, and perhaps HTML depending on the use-case.
Where this becomes more important is with other context nodes, for example, with a P element. When setting inner HTML on a P element and encountering </p>, the fragment parser should create an empty P element child of the P (even though this produces an invalid DOM). With the current behavior, it would detect the open P on the stack of open elements and close it, skipping over the expected empty invalid P.
A fix should involve removing the context node from the stack of open elements, and it may adjust how breadcrumbs are reported. But should breadcrumbs show the root node (HTML) or only show the nodes inside the context? There's no direct analogue in JS since access is only available once the parse is complete.
Based on the working operation of the fragment parser, it seems like the breadcrumbs for <body> context should start at the top-level nodes within that BODY element, not at HTML. This brings two drawbacks:
- This would be a breaking change in the HTML Processor (though it's unlikely much or any code in practice relies on this behavior, since
matches_breadcrumbs()matches partial breadcrumbs). - We lose the easy opportunity to verify the full breadcrumb path in
matches_breadcrumbs()by starting atHTML. Now, it would also be necessary to confirm the length of the breadcrumbs or add a new placeholder tomatches_breadcrumbs()to indicate the end. e.g.matches_breadcrumbs( array( 'root', 'P' ) )to match aPelement at the top level.