According to the dataset specification, how is element.dataset meant to delete data attributes? Consider:

<p id="example" data-a="string a" data-b="string b"></p>

If you do this:

var elem = document.querySelector('#example');

elem.dataset.a = null;

elem.dataset.b = undefined;

elem.dataset.c = false;

elem.dataset.d = 3;

elem.dataset.e = [1, 2, 3];

elem.dataset.f = {prop: 'value'};

elem.dataset.g = JSON.stringify({prop: 'value'});

the DOM becomes this in Chrome and Firefox:

<p id="example" 

   data-a="null" 

   data-b="undefined" 

   data-c="false" 

   data-d="3" 

   data-e="1,2,3" 

   data.f="[object Object]" 

   data.g="{"prop":"value"}"

></p>

The Chrome/Firefox implementations mimic setAttribute. It basically applies .toString() first. This makes sense to me except for the treatment of null because I would expect that null would remove the attribute. Otherwise how does the dataset API do the equivalent of:

elem.removeAttribute('data-a');

And what about boolean attributes:

<p data-something> is equivalent to <p data-something=""> Hmm.