Should <link rel="stylesheet"> work inside shadow DOM?

[domenic]

Right now it uses "in a document", meaning we are not sure whether it should be "in a document tree" or "connected": https://html.spec.whatwg.org/multipage/semantics.html#link-type-stylesheet:concept-link-obtain

I think we should allow this (i.e. switch to "connected"). If we allow <style>, why not <link>?

[rniwa]

Also see #526

[hayatoito]

• link is banned in a shadow tree.
  See http://w3c.github.io/webcomponents/spec/shadow/#inertness-of-html-elements-in-a-shadow-tree
  We can replace "in a document" with "in a document tree" for link.
• Regarding style, yeah, see ShadowRoot.styleSheets should return an empty StyleSheetList if the shadow root is not connected #526. The current blink's implementation: we allow <style> if and only if it is connected.

[rniwa]

I don't think restricting link element to the document tree makes much sense. We should allow it in shadow trees.

@travisleithead @annevk

[rniwa]

On my second thought, I do recall a discussion about not allowing loading of external scripts and stylesheets to discourage people from loading the same styles/scripts multiple times in a single page.

[annevk]

We already decided to allow scripts. I'm fine with allowing this as well. Presumably long-term this will mostly be done through an API.

[rniwa]

On the other hand, it might be wise to not allow stylesheets or scripts to be loaded in shadow tree until that we figure out what we want to do with the new thing though. Otherwise, we may end up tying ourselves to some undesirable constraints.

[TakayoshiKochi]

As @annevk already pointed out, in long-term we would use constructable stylesheets
http://tabatkins.github.io/specs/construct-stylesheets/.

Just for historical reference, some discussions related to this:

• https://bugs.chromium.org/p/chromium/issues/detail?id=313458
• https://www.w3.org/Bugs/Public/show_bug.cgi?id=22539
• [Shadow]: Add support for external stylesheets using link rel="stylesheet" (bugzilla: 22539) #82

[domenic]

It seems like the argument against is that UAs are able to optimize <style> with the same text content appearing many times in the same document, but are not able to optimize <link rel="stylesheet"> with the same href.

Is that completely true? Even if the response has appropriate long-lived caching headers? I guess on initial page load with an umprimed cache it would probably still make a bunch of requests, which is bad?

[hayatoito]

Yeah, I think UA can optimize it in theory, but such an optimization is not guaranteed by the spec. As far as I remember, that was a concern from web developers.

Thus, we want to provide a standard way where such an optimization is strongly guaranteed.

[domenic]

Hmm. I don't find that compelling, because as a web developer it seems like the same argument applies to multiple <style>s. That is, in theory it can be optimized so that the UA does not parse it multiple times or do other work, but in practice it is not guaranteed. (I think Apple has in fact stated they have such an optimization ready, whereas Google is pushing constructible stylesheets because they do not, if I recall previous discussions.)

Maybe it is an issue of degree of badness? Multi-parsing is not as bad as multi-fetching?

In general any asymmetry between <style> and <link rel="stylesheet"> is really surprising for web devs IMO...

[annevk]

Yeah, and given <style>@import ...</style> it doesn't make sense to support one and not the other. We should just allow both.

[hayatoito]

Thanks. I do not have a strong opinion on this issue, historically. :)
I am totally okay to allow <link rel="stylesheet"> if it is connected, hoping that web developers would use it in a good manner.

[trusktr]

It is definitely intuitive that <link> should work where <style> does. By the way, why not <style src=""> similar to script tags (seems even more intuitive)?

[hayatoito]

Can we agree that <link rel="stylesheet"> should work if and only if it is connected?
I think it is reasonable, given that we have already allowed <style>@import ...</style>, which developers have been using as a loophole, I guess.

If we can agree on that, I will send "Intent to implement and ship: Allow <link rel="stylesheet"> which is connected" to blink-dev.

[trusktr]

Can we agree that should work if and only if it is connected?

Yes, and if it is disconnected then connected into another shadow tree then the styling should move there too.

[rniwa]

I don't understand why we want to restrict to the connected shadow trees. What's the rationale?

Given style should apply to elements inside a disconnected shadow tree, and that forces @import to work, it seems rather inconsistent for link to not load in a disconnected shadow tree.

[rniwa]

Also see the issue #526.

[trusktr]

@rniwa

Can we agree that should work if and only if it is connected?

I thought the conversation was about <link> being connected inside the shadow tree, not about the shadow tree being connected. Maybe it should wait for both to be true (makes sense): link element connected and shadow tree in a connected ancestor (because a shadow tree is always connected to it's host even if the host is in a disconnected tree, right?).

[domenic]

No, a connected restriction here makes no sense... I don't understand why it's being discussed.

[domenic]

That is, I think this should be governed by the normal rules for stylesheet links, with "in a document" replaced by "connected" to allow it to work inside shadow trees, and not worrying about its root's connected status, or the element being disconnected, or somehow "moving styles" whatever that means... or anything like that at all. This seems fairly simple... We should just spec it.

[trusktr]

or the element being disconnected and moved

What should happen if the <link> is moved to another shadow tree?

[domenic]

This is all covered by the CSS scoping spec.

[trusktr]

Gotcha, so I assume exactly the same as with <style>.

But, <link> has a network aspect. If the <link>is connected to a document fragment, should it still fetch the resource? Or wait until inserted into real DOM?

[trusktr]

I meant, if the <link> is connected to a shadow root inside a document fragment.

[domenic]

Please read the above link I gave, mentally replacing "in a document" with "connected".

(You may also want to re-read the definition of connected, as the way you're using it doesn't seem to match any existing definition.)

[trusktr]

From earlier,

Can we agree that should work if and only if it is connected?

It's not clear what parts of <link> working we're even talking about.

Maybe it should just do everything it already does now (nothing different), and waiting to apply styles until it is connected, because only then can the we possibly know to scope style to a shadow tree.

[domenic]

I think it's pretty clear, between HTML and CSS Scoping. If you disagree, I'd suggest waiting until we get a spec finished and then you can investigate in more detail.

[domenic]

I think that's generally intentional. At least, that's how we do things in the rest of the platform, for images, videos, scripts, etc., and how link works today.

[rniwa]

Well, for document, yes. What we're discussing is what we're gonna do with shadow DOM, so there is no intention whatsoever. We're defining what that intended behavior would be here.

[domenic]

Well, we already made the decision for a few of those (notably scripts), to just change "in a document" to "connected". In no case have we decided to start loading resources when the resource's shadow-including root is not a document.

It seems important, for example, that if you wrap a script or link or image or... in a shadow tree, it doesn't suddenly start making resource fetches. E.g. a custom element wrapping an image (like x-gif) should behave just like an image and not perform fetches until the custom element becomes connected (and thus the img in its shadow DOM becomes connected).

[rniwa]

@bzbarsky @travisleithead @smaug----

[annevk]

I agree with @domenic. I don't think "in a document" should become "in a document" or "whose root is a shadow root" generally. "Connected" makes sense.

[smaug----]

Hmm, making to work when not connected? that sounds very surprising.
Since has had the requirement for the element to be in document before loading, that is what I'd expect for shadow DOM too (that element is connected to the document). That would give some consistency.

Would be super weird if creating a random (not in document) DOM subtree and link in it, wouldn't load the stylesheet, but then using that same subtree, on some element in it do el.shadowRoot.innerHTML = "<link rel=stylesheet href=..>"; would actually load the stylesheet.

So, yes, we need connected-ness here.

[rniwa]

Okay, let's go with "connected" then.

[ghost]

Sorry if I’m misunderstanding something, but I think that only loading when connected could cause unexpected behavior if someone tries to inspect the computed styles of the elements inside the shadow tree.

[annevk]

@Zambonifofex could you elaborate?

[ghost]

Well, if I have the following stylesheet being linked:

button
{
    color: rgb(12, 34, 56);
}

…and but is a button inside the shadow tree, I’d expect the following to always log "rgb(12, 34, 56)", even if the shadow tree isn’t in a document:

console.log(window.getComputedStyle(but).color);

(by the way, sorry if I’m using wrong terminology)

[domenic]

Why would you expect that? It's not true for node trees today. Why would you expect the result to be different when the disconnected node tree is rooted at a shadow root, instead of being rooted at e.g. a <div> or a DocumentFragment?

[ghost]

I guess I feel that way because shadow roots limit the scope of the stylesheet just like documents do. I’d expect the same if the link is inside a document that isn’t being displayed to the user (i.e. one created by document.implementation.createDocument/createHTMLDocument).

[domenic]

Ah, no, stylesheets are not fetched in such documents. (See whatwg/html#1516 and http://software.hixie.ch/utilities/js/live-dom-viewer/?saved=4299)

[ghost]

@domenic then I guess it makes more sense to require “connectedness” to fetch stylesheets. Still, that’s not what I’d expect at all.

[trusktr]

It seems important, for example, that if you wrap a script or link or image or... in a shadow root, it doesn't suddenly start making resource fetches.

That sounds perfect.

[hayatoito]

+1 for the current conclusion.

Thank you, folks. I am totally fine: <link ref=stylesheet href="..."> fetches a stylesheet if and only if it is connected (except for documents without browsing contexts).
It looks what I would like to say was already covered by others. :)

[domenic]

I will try to spec this as soon as someone (cough, @annevk) reviews whatwg/html#1516

[olanod]

Hello! I have a couple of issues described in this SO question regarding FOUC and measurement of elements while a stylesheet being loaded, is anyone here willing to share some thoughts about it?