Improvements to namespaces and attributes in serialization

[fangel]

Hi,

I started out by adding test-cases for the serialisation, and found a few issues once I compared to how browsers (ehh, phantomjs) does it. I found a few discrepancies that I have addressed. Additionally I ensured that elem.[attr] is equivalent to elem.getAttribute("[attr]").

So the new or fixed features are:

• elem.[attr] is equivalent to elem.getAttribute("[attr]")
• elem.style.backgroundColor and other CSS properties that include hyphens are now serialised correctly
• Namespaced attributes are now serialised correctly (I implemented it wrong in the previous PR)
• Void-elements are now serialised as void-elements - ie, they do not have an end-tag. This is only done if the documents namespace is the HTML namespace.

Known differences between min-document and browsers:

• document.createElement('input').getAttribute('type') is null in browsers, but text in min-document. This is because in both browsers and min-document, document.createElement('input').type is text - ie, in browsers there is a difference between .type and .getAttribute('type');.
• document.createElement('input') is serialised as <input> in browsers, and <input type="text"> in min-document. This is the same issue as the previous discrepancy.

Kind regards
Morten

[Raynos]

elem.[attr] is equivalent to elem.getAttribute("[attr]")

that's not how the DOM works. elem.[propertyName] and elem.getAttribute('[attrName]') are completely different.

[Raynos]

We cannot write to this. We must write to this._attributes['default-namespace'] or something. We can even have this._attributes and this._nsAttributes

[Raynos]

Other then props vs attributes the rest looks good to me.

[fangel]

Well, at least for HTML documents, and supported attributes is does.

E.g

el = document.createElement('div')
el.id = "foo"
el.getAttribute("id") === "foo" // <- true

I considered having a white-list of attributes that if the namespaceURI was the html-namespace, then have this behaviour, but I figured it would start pushing this project away from being a minimal DOM implementation.

But if you don't want this behaviour, then I'll remove that commit, add a commit that fixes serialisation of things like

el = document.createElement('div')
el.foo = "bar"
el.setAttribute("foo", "baz")

and force-push to this feature-branch to update the PR. The problem with the code above is that it would serialise to <div foo="bar" foo="bas"></div>. Do you have any input on what this should serialise to?

-M.

[Raynos]

@fangel

The problem you are running into is the need to have a property -> attribute mapping. Because x.className serializes as <x class=...>

You also need deduping, i.e. if an attribute is written once dont write it again.

I recommend:

• change the serializer and add a deduper to not double serialize props and attributes
• ignore the x.id = 'foo'; x.getAttribute('id') edge case. Anyone setting with props and reading with attributes deserves what they get.

However let's not simplify the problem down to "attributes and properties are the same" that is a wrong implementation. I'd rather have an incomplete or subset implementation then a wrong one.