Should <template for> patch when shadowrootmode doesn't attach a shadow root? #13075

What is the issue with the HTML Standard?

The template start tag steps in "in head" give a valid shadowrootmode precedence over for, even when no shadow root gets attached:

If template's shadowrootmode attribute is not in the None state:

1. If any of the following are true:

   • the parser's allow declarative shadow roots is false;
   • hostIsRoot is true; or
   • the parser's insertion target redirection map[declarativeShadowHostElement] exists,

   then insert an element at the adjusted insertion location with template, insertionLocation, and true, and return.

...

Otherwise, if template has a for attribute: ...

Attaching the shadow root can also fail, in which case the steps likewise insert template and return. So per spec, for is ignored whenever shadowrootmode is valid.

Chrome Canary 157 patches instead whenever no shadow root is attached:

const div = document.createElement("div");
div.innerHTML = '<?marker name="x"><template shadowrootmode=open for=x>A</template>';
div.innerHTML;
// Spec: '<?marker name="x"?><template shadowrootmode="open" for="x">A</template>'
// Chrome Canary 157: 'A'

The same happens with a top-level template in setHTMLUnsafe(), and with Document.parseHTMLUnsafe() when the host already has a shadow root. Chrome only skips patching when the shadow root is attached. Gecko's in-progress implementation (bug 2053472) does the same as Chrome. template-for-superseded-by-shadowrootmode.html only covers the case where the shadow root is attached.

Maybe the spec should match: fall through to the for steps when no shadow root is attached, instead of returning.

cc @foolip @noamr