formula.js: consider leaving tags in the document

[pkra]

Via https://github.com/AmerMathSoc/ams-html/issues/938

In #422 we added extraction into data attributes. However, it might be easier and better to "just" keep the tags in the document.

[pkra]

Note: one problem will be that formulas inside tag would be turned into $...$ which is what we'd want to avoid.

[pkra]

If we make the element hidden, we could keep the old approach and remove it later. That is, avoid a breaking change.

[pkra]

Some notes from an experimental implementations

Since ams-eqn-store reads the innerHTML of formula.js "output", we need to either put the tags outside of it or change ams-eqn-store.

I tried to just an element with the tags after which didn't go well on the ams-html side -- the math panel creation has trouble because it is working with cloned nodes, so the sibling is tricky to find. Working around it temporarily, I ran into odd errors I couldn't track down - at which point I decide to stop because it was getting too hacky.

So this seems to require a somewhat larger change throughout the toolchain. It might be that some things get simpler in the long run (e.g., making formula elements plain wrappers and having tex-math and tags as children). But it needs more investigating.

[pkra]

Here's the diff of where I had left things on this end:

diff --git a/lib/elements/formula.js b/lib/elements/formula.js
index dfb3274..22a43a4 100644
--- a/lib/elements/formula.js
+++ b/lib/elements/formula.js
@@ -12,12 +12,15 @@ export default function (htmlParentNode, xmlnode) {
     htmlParentNode.insertAdjacentText('beforeend', '$');
     return;
   }
+  const hasLinkedTag = xmlnode.querySelector('target tag');
+  const tagContainer = `<div hidden data-ams-doc="tags"></div>`; //NOTE we store copies of tags for easier re-use downstream; cf. tags.js 
   if (xmlnode.getAttribute('content-type') === 'text') {
     const div = this.createNode('div', '', {
       'data-ams-doc': `math text`
     });
     htmlParentNode.appendChild(div);
     this.passThrough(div, xmlnode);
+    if (hasLinkedTag) div.insertAdjacentHTML('afterbegin', tagContainer)
     return
   }
   // Otherwise
@@ -28,6 +31,7 @@ export default function (htmlParentNode, xmlnode) {
   htmlParentNode.appendChild(span);
   if (mathMode === 'block' && xmlnode.querySelector('tex-math[has-qed-box]'))
     span.setAttribute('data-ams-qed-box', 'true');
+  if (hasLinkedTag) span.insertAdjacentHTML('afterend', tagContainer);
   this.passThrough(span, xmlnode);
   // NOTE we (sometimes?) get extra whitespace from childnodes; needs test
   const text = span.innerHTML;
diff --git a/lib/elements/tag.js b/lib/elements/tag.js
index 4a0545b..397c31c 100644
--- a/lib/elements/tag.js
+++ b/lib/elements/tag.js
@@ -27,4 +27,8 @@ export default function (htmlParentNode, xmlnode) {
         if (xmlnode.getAttribute('parens') === 'yes') span.insertAdjacentText('beforeend', ')')
         htmlParentNode.prepend(span);
     }
+    // tag extraction 
+    // if some tag in the equation is linked, we store all tags in a special element in the HTML parent
+    const tagsElement = htmlParentNode.closest('[data-ams-doc~="math"]').nextElementSibling;
+    if (tagsElement?.getAttribute('data-ams-doc') === 'tags') tagsElement.insertAdjacentHTML('beforeend', `<span>${this.passthroughIntoHTMLString(xmlnode.cloneNode(true))}</span>`);
 };

[pkra]

Here's a radical thought: let's preserve (outer) tex-math, i.e., instead of

<span data-ams-doc="math block">\tag{$x$}</span>

something like

<span data-ams-doc="math block">
  <tex-math>\tag{$x$}</tex-math>
  <span hidden data-ams-doc="tags">
    <span>
       <span data-ams-doc="math inline">
         <tex-math>x</tex-math>
      </span>
    </span>
  </span>
</span>
``

(which is a bit convoluted when there's math in the tag but still)

[pkra]

Fun fact: we don't handle "text equations" properly in ams-html (createMath doesn't pick up on them).

[pkra]

We don't have to collect tags in text equations because their tags come out as spans with inline math already (well, if we had any with inline math in their tags).

[pkra]

This seems to be working out ok downstream

• fairly small changes in store prep and injection (to adjust to tex-math elements having the actual TeX)
• small change in ams-html, switching to pick tags from the element with tags instead of the attribute - no visual changes

I haven't looked into all test failures here - a ton of changes are expected and I'd like to take the opportunity to clean up the formula-related tests.