postprocessing can "change" the skeleton node (if it's a link)

[pkra]

I had filed Speech-Rule-Engine/speech-rule-engine#801 but it turns out it's sre2tree's fault.

The use case was \href{a}{b, c}.

While sre2tree handles the skeleton node not being the root here:

sre-to-tree/lib.js

Lines 124 to 125 in 48503b2

	export const sre2tree = (node) => {
	const skeletonNode = node.querySelector('[data-semantic-id]:not([data-semantic-parent])');

in this case it (re)moves the relevant attributes in

sre-to-tree/lib.js

Lines 100 to 111 in 48503b2

	// rewrite (proper) links
	// NOTE. Since MathJax does not support <a> in internal format, the SRE structure is placed inside the a. We assume it's on the first semantic child
	const firstSemanticChild = child.querySelector('[data-semantic-speech]');
	if (!firstSemanticChild) {
	console.warn('Link without semantic child');
	return;
	}
	// move all attributes
	[...firstSemanticChild.attributes].forEach(attr => {
	child.setAttribute(attr.name, attr.value);
	firstSemanticChild.removeAttribute(attr.name);
	})

(since the anchor is the parent of the skeleton node)

So when we get to

sre-to-tree/lib.js

Lines 136 to 143 in 48503b2

	if (node !== skeletonNode) {
	moveAttribute(skeletonNode, node, 'data-owns');
	moveAttribute(skeletonNode, node, 'aria-label');
	moveAttribute(skeletonNode, node, 'aria-braillelabel');
	moveAttribute(skeletonNode, node, 'aria-level');
	skeletonNode.removeAttribute('aria-hidden');
	skeletonNode.setAttribute('role', 'presentation');
	}

the skeleton node no longer has the attributes we're trying to move.

[pkra]

The trouble is: in this case, if we move the attributes up to the root, we lose the link functionality.

[pkra]

Looking for uses case in AMS publications, I find merely 11 cases across 7 files. There are many more cases of a "wrapping link" situation, but those have no real tree structure (which technically leads to another bug elsewhere as those are turned left as role=img, also losing the link).

In all of those cases, these usually are very short link texts (I could not find one that wasn't a simple reference to a labelled equation).

In addition, there's a question whether texml-to-html should be wrapping xref(-group?) content in \text{}.)

While I'm not sure if these few cases are primarily because of files having been rendered with on older versions of SRE (i.e., more of the "non-tree"s might turn into trees).

Given the small amount of real world use cases but especially the low complexity of non-visual presentation, I'm leaning towards a hack downstream (turning such "trees" into "flat" links).

[pkra]

Good progress on this today. But downstream integration is surfacing some more subtleties I need to look into.

[pkra]

This is really a new feature: supporting edge cases in SRE output where tree structures cause problems.