"NotFoundError" might be suitable? It's used for TextTrack-TextTrackCue ownership check.

TypeError is nicer if we ever want to abstract this into IDL.

I don't think we could feasibly put this into the IDL type system... I'm reasonably convinced by the symmetry with the existing ownership check. I'll switch.

I guess we somehow know that reentrancy is not a problem here? Does it rely on the protections added for formdata?

Hmm. I guess this does open up the possibility of

form.onsubmit = () => { form.requestSubmit(); }
form.requestSubmit();

We could add a flag to protect against that. I guess we should do so, for symmetry with reset(). Good catch.

Ping @tkent-google to check my logic, and also update the tests.

We should have the reentrancy check in form submission algorithm, not in requestSubmit()

click() of a submit button already has a reentrancy issue.

form.onsubmit = () => {
  form.querySelector('[type=submit]').click();
}

When a submit button owned by form is clicked, submit event is dispatched, and click() invokes another form submission.

Chrome, Safari, and Firefox already have a reentrancy check for this pattern, and the form submission invoked by click() is aborted.
On the other hand, form.submit() in submit event handler should be allowed for compatibility.

WebKit/Blink implementation is something like:

• <form> has in-submit-event flag, initially false.
• It should be set to true before firing submit event, and reset to false after it.
• Add "If the submitted from() method flag is not set and in-submit-event flag is true, then return." before step 6 of form submission algorithm.

Thank you for catching that deeper problem. Do you think this flag might be mergeable with the "constructing entry list" flag?

Either way, we should fix that missing check in a separate PR, and rebase this PR on top of it. I am heading to sleep now, but maybe you could work on a spec and tests PR for it? Or I can do it tomorrow.

Filed #4620, and
made a specification PR; #4621