WebApps' Web Components f2f Meeting

21 Jun 2013


See also: IRC log


Edward_OConnor, Ryosuke_Niwa, Bear_Travis, Divya_Manian, Alex_Komoroske, David_Baron, Priyank_Singhal, Steve_Orvell, Daniel_Freedman, Dimitri_Glazkov, Elliott_Sprehn, Scott_Miles, Tab_Atkins


<jkomoros> ScribeNick: jkomoros

<ArtB> ScribeNick: jkomoros

<ArtB> Scribe: Alex

<scribe> Chair: Dimitri_Glazkov

<scribe> Meeting: Styling Issues in Shadow DOM and CSS

Quick Overview of Shadow DOM concepts

<divya> i think we need cofffee

DG: Hoping that the main focus of this meeting will be primarily arounds CSS + Shadow DOM
... we had one original idea, but developers trying to use it gave feedback that it wasn't exactly the right "knobs"
... there are people here who are "Browser Vendors", and there are people who are the "web developers"
... a bunch of folks in the latter group here are from Polymer, Daniel Buchner (who should join at some point) represents x-tags
... and then spec folks, fantasai and tabatkins
... who aren't here yet.

dbaron: Blake Kaplan and William Chen (?) have been working on Shadow DOM at Mozilla
... and I've been talking with them

[by the way, we all took a coffee break]

[break over]

DG: The general idea of Shadow DOM is that you have an ability to create trees, like before, but connected for rendering purposes, render in place of nodes in document

<rniwa> wasn't there explainer somewhere?

DG: this existed in many different systems before. It allows composability (one tree vs multiple)

<rniwa> is https://dvcs.w3.org/hg/webcomponents/raw-file/57f8cfc4a7dc/explainer/index.html still up to date?

DG: if I can replace the rendering of a node, what happens to its children?

<slightlyoff> rniwa: or, also: http://glazkov.com/2011/01/14/what-the-heck-is-shadow-dom/

DG: the general overview gets trickier and trickier, but we have converged on a solution in today's Shadow DOM spec

[dglazkov draws a diagram on the board]

<rniwa> Also see: https://dvcs.w3.org/hg/webcomponents/raw-file/57f8cfc4a7dc/explainer/

scribe: every node that has children, you can associate (off to the right) with a shadowRoot: a DocumentFragment with extra stuff in it

<slightlyoff> rniwa: this loads for me: https://dvcs.w3.org/hg/webcomponents/raw-file/57f8cfc4a7dc/explainer/index.html

scribe: extra stuff is effectively a subclass of DocumentFragment. Things like getElementByID, querySelector. Stuff that has migrated into Document mainly anyway

dbaron: So those just query what's in the Shadow DOM?

<rniwa> slightlyoff: oh oops, yeah. i guess it doesn't have an ordinary index.html > / rewrite :/

DG: think of the line connecting ShadowRoot is not a normal connection--it's a separate tree
... insertion points can be any elements inside the tree. They're called <content>
... we use a rhombus for insertion points
... <content> name comes from XBL2
... you can have more than 1 content
... content can have a select attribute, which takes a narrow subset of CSS selctors
... that match against children of the parent node.
... currently limited to ID, tagname, attributes, and class
... no combinators.
... that's the conceptual model. But actually a node can have MULTIPLE shadow roots
... the method on the ndoe is "createShadowRoot"
... there's an ordering.
... Sometimes the element already has a shadowtree (like InputElement or TextArea)
... they're basically the same as how the native implementation might be done
... it's actually a stack of trees. new ones go on top of old ones; the newest one is the visible one. The ones underneath don't render
... there's a concept of older and younger shadow tree
... youngest one is the one that gets rendered
... soemtimes you want to use parts of the older shadow tree
... which is why there's an insertion point called <shadow>
... when you put it in a shadow root, it will show whatever what is the older shadow root
... it allows the youngest guy to channel the older guy
... explicit children can only go to one insertion point.
... there's an idea conceived by Jan on the polymer-dev list, the shadow acting as a function (?)
... but as of now, there is an order, only selected once
... this allows developers to take existing elements, and adorn them with existing stuff from older shadow trees
... if there's nothing in the older shadow tree, it works as the last content element--whatever hasn't
... been distributed
... whole point of Shadow DOM spec is distributed. That's the majority of the spec
... how are they distributed, what's the effect
... things like focus, events, and rendering/sytling
... the latter is what I want to talk abou ttoday
... the others we have figured out mostly

dbaron: I was involved in the XBL RCC thing in 2004 (?) so these concepts are not all new to me

DG: now we get into style
... this is where things get interesting

<dbaron> (also XBL1 :-)

DG: if the shadow root is a document fragment, what does that mean from styling perspective?
... if I'm distributing a text node into a content element, what is its style?

rniwa: What does the current spec say about style?

[I missed about 30 seconds :-( ]

dbaron: I think it's worth separting selector matching and inheritance

[esprehn draws a diagram on the board]

es: When you attach the shadow root, content doesn't render. But in this shadow, the content is "teleported" as though it was there when rendering
... so you get styles from where you came from, and styles where you're going
... there's a way to reset styles at shadow boundary

<divya> http://s3.amazonaws.com/Gyazo/1371838156.png

es: when the tree gets flattened out, conceptually it gets flattened out

[es draws the "composed" result on the board for clarity]

scribe: we use "composed" tree to mean, the thing with all of the things teleported
... don't use "flattened" tree

dg: Although at some point we might, depending on if there's mutiple trees

rniwa: If you have a style in the distributed content, that follows hierarchy in original content
... and merge in with shadow styles
... I'm not sure that even in an complex widget that makes sense
... things get really wonky

sorvell: It's just inherited styles that work this way

es: One special case si that if you have a style element inside of the shadow root, it's automatically scoped to the shadow root

dbaron: So the selectors in the scoped style only match things in the shadow tree, NOT stuff that gets "teleported" there

sorvell: This is one part of the spec as a developer that makes total sense
... allows you to worry just about this shadow root. it works really well in practice

sjmiles: Occasionally you have to pierce through that barrier, that's when it gets harder

<divya> http://s3.amazonaws.com/Gyazo/1371838492.png

sjmiles: as a practical matter WE haven't run into that problem
... (confused styles)

dbaron: By pierce through, do you mean that someimes you want the explicit children of the node to inherit from what before as opposed to from shadow dom?
... is there a way to say that, in that case, the span shoul dinherit font size but not color

sjmiles: no, it's all or nothing
... (basically)
... we haven't run into that need in practice y et
... that's based on empirical data with n points, where n is a relatively small number in the grand scheme of things
... it's possible at some point in the future someone will need it, but we don't now

dg: let's enumerate the cool styling hooks that we have today, then figure out which ones are missing
... 1) Style scoped.
... it's acgtually a close cousin of shadowRoot. It's very similar scoping behavior
... but it's a scoping NODE, and style scoped is a scoping ELEMENT
... but they have similar abilities, except none of the styles from the document (outside the SR) don't apply down.
... style scoped in isolation, essentially

dbaron: So nothing from the author style sheets don't match SR. But still UA styles

DG: we have applyAuthorSTyles
... that allows the component to explicitly allow outside styles out to come inside
... user styles are treated like Author styles

eo: It's problematic that user styles by default get blocked

dg: Actually, we don't know what we do here, we need to check

tabatkins: It's reasonabl eto say, yeah, User styles apply by default

dbaron: How selector matching works is intersting

dg: If you say applyAuthorStyles, there's still a weird relationship, where you even though a child of shadowRoot might LOOK like it's child of the host
... it's actually not. The selector matching can either be fully in the host, or in the SR

[I think I got that right?]

scribe: so in this example, div > div will not match
... but if you just do `div` and have applyAuthorStyles it will match both divs

sjmiles: If I put div class=foo, and foo is defined in document, it won't see that
... as a user I go, applyAuthorStyles will make it work, but it won't

dg: No, that will work

es; BAsically, the selector must COMPLETELY Match outside, or completely inside. There's no boundary crossing

eo: What about a boundary crossing combinator?

<dbaron> db: yeah

dg: That's what we want to talk about today: :-)
... There's another flag on SR that says resetStyleInheritance
... it's very powerful
... everything inside of the SR, when you flip to true, it will look like it's initial styling

dbaron: Kind of like you had a parent in between all:initial

dg: similar thing exists on insertion points
... so that you can have the styles in SR not go into the composed children
... that's all the styling machinery (minus any boundary-piercing things)
... but this isn't enough. how do you make the subtrees intreact with doc?
... similarly, sorvell wants to be able to style inside the shadow tree, style the composed children as well
... like say in a tab strip, styling the active children
... you want to be able to let SOME stuff in from SR in from document, and also in
... similar for content

dbaron: The XBL solution to that is that you have a separate binding for active tab that is different, point to that instead

dg: We didn't want the content of the host to not know what's happening to it (?)
... we had two solutions, both of which have strengths and weaknesses
... 1) let CSS Variables bleed through the SR boundary. So you could specify a CSS variable in doc, and catch it inside the SR

tabatkins: in Style WG, we decided that variable resetting isn't covered in "all". YOu'd need to explicitly say "vars" as well (syntax I probably got wrong)

dbaron: i don't like describing inheritance blocking as all property

fantasai: You probably do want the ability to jump inheritance over the shadows

dbaron: I'm nervous about cutting off inheritance from stuff outside SR to stuff inside. I'm less nervous about inheriting from the shadow into the children

tabatkins; that turns out to be extremely popular for writing components in the real world

scribe: like components in jquery have to go through and manually reset everything
... they want consistent, predictable starting point--even if they allow poking in after that

fantasai: But imagine we're using this to rearrange list items into new structure. The expectation of the author is that setting font on the root of the doc, it sets it everywhere. But if you do cut off inheritance, then those list items will have UA default

tabatkins: That's why it's a flag. Component authors can decide if it works

es: Actually, default is to allowing

sjmiles: So if you turn it off, the component author did it on purpose

dbaron: So in the cases where you have a binding wiht lots of content inside, like say a tab widget. You probably want the inheritance through to your big piece of content. Bu there's some little content you don't want it

dg: Think about disqus use case. They mostly want it to match the blog they're embedded in. But if you're building an app, you might want a certain style that's very particular no matter where it is
... like the G+ share widget, as an example, that wants to have complete control over exactly what's inside it

sjmiles: db makes a good point

tabatkins: wihtin the shadow, if you want to block it only in some places, the 'all' property exists

sjmiles: Or make a component for just the parts where you want to reset it

sorvell: We don't use resetting much in practice, it's such a blunt tool
... generally we want to control a small number of properties

rniwa: In disqus use case, you want to be able to read the background color of surrounding, but decide how to interpet that

tabatkins: Today you can do that. Use 'all' to reset all, then 'inherit' for the other properties you want to allow in. Or the other way around, use 'initial'

es: LIke in a facebook button as an example, you want to force the font, but don't care about the size of it

dg: Let's keep going with explaining the tools
... those variables are cool but not enough
... we see this alot with WebKit's internal input elements
... you want access to a sepcific element to style arbitrarily
... leads to:
... 2) Custom pseudo elements
... you define a pseudo attribute on an element
... then you can use it with standard ::foo syntax in selectors
... so like <outer-element>::x-foo

dbaron: Like functionality, but want a function

eo: agreed

dg: Agreed, at the time when we proposed this dave hyatt didn't want it to be a functional syntax, but we can revisit

es: ONe of the goals of the project is to explain lower-level magic

eo: I don't agree with that goal, for what it's worth

es: If we swithc to functional syntax, we miss out on explaining the ::foo magic

rniwa: We could change the syntax for pseudos like that if we want, only blink and webkit do this

dbaron: If implementations want to implement web platform features that have pseudos, they can have their own versions that don't use the functional syntax
... I'm SLIGHTLY sympathetic to wanting to explain the magic. But some of them are things we really don't want to freeze
... like if we had done styling of form controls "right" back in 2000/2002, it wouldn't have been web-compatible to make the form controls used on iOS and ANdroid
... because the web would have depended on fixed structures that work on destop, but don't make sense on mobile devices

dg: There is a larger debate here. I want to table that for now. Keep it in mind today, but avoid engaging today

eo: We'll only be able to make so much progress without it

dg: So even with this second knob, it doesn't complete all of the use cases from developers

<dbaron> (I think he used "put it aside" rather than "table" (which is en-GB/en-US ambiguous).)

dg: now I'm crossing threshold to the boundary-crossing thing
... I'll first describe things they way they WERE/ARE

divya: What do you mean by "functional syntax"?

dg: things like ::shadow(foo)

dbaron: the advantage is, there's a rule in selector spec that says rules UA doesn't understand get dropped
... pseudo elements/classes are part of that rule. WebKit/Blink don't do that correctly
... all other browsers drop the entire rule, but WebKit/Blink retains those

es: That was willful; we can fix it

tabatkins: But peopel do use it today already :-(

dg: In querySelector, incidetnally, we don't violate spec
... onto the new things that we're thinking of
... in order to select things that are distirbuted into an insertion point, we invested the distirbuted pseudo element function
...: :distributed(---------)
... where ----- allows combinators
... on an insertion point, inside of a SR, it matches the element that was distributed into that matches the inner selector in the function

es: example: content::distributed(span) { border: ________ }
... in the example we diagrammed, that style that earlier didn't match, now matches

tabatkins: Remember, this is current junky stuff that we don't like

es: Essentially content has a list of things that have been distributed in, the selector inside the parens says which in that list to select

sorvell: It's relative

dg: It's relative to the virtual node that represents the thing that envelopes all distributed elements

sorvell: use case: i want to style all children, not all descendants
... so you can do like content::distributed( > span)

es: It's like find() on content

dbaron: I don't know if I like leading cominbators yet

fantasai: I have reservations, but I think at this point we have to go with it, everyone expects it to work that way

tabatkins: jquery uses it for years, and now it's documented in selectors level 4. It's a small section

dbaron: leading combinators only work when there's an implicit node being targeted to

sjmiles: This is very necessary in our experience

divya: Can I have a class on the content element and use that in the selector?

dg: yes

es: Although content element itself is NOT stylable
... which I don't like. I wish that I could style the content to, say, display:none it
... currently it has no effect
... it's bizarre

dg: I agree

<stearns> +1 to styling content nodes

dg: I would like to explain content as a display:contents

es: In the current model, it's easy to distribute two things, but if you want to hide it, you need ANOTHER wrapper
... styles targeted at <content> don't inherit down; it's unstylable, no rendernode

fantasai: If it's an intermediary, it makes for example uls and lis nested not work

dg: I hope we can solve it by having <content> have display:contents on it

fantasai: but that doesn't address the ul/li use case

<fantasai> also mentioned :nth-child

<fantasai> ul/li shouldn't be a problem

dg: let's talk about @host

<fantasai> otherwise

dg: we want to get rid of this
... when you put a SR inside a tree, I want to be able to apply borders on the component, for example
... works like @host { [selector] { border: 1px solid red }}
... the inner selector matches only host

dbaron: and what would antyhing other than *

eo: You can imagine a case where you want to embed a widget in two different places, and you only want one (regarding why you'd want something other than *)

tabatkins: And because of is attribute, you could have one component with different tag-names

dbaron: So @host lets the SR influence the containing box
... I think that in XBL1 we replace the outer box, but I might be misremembering

dg: This is the entire family of styling stuff. Now we want to get rid of many of these

<fantasai> side discussion of pseudo-element syntax, vs . combinators vs. @rule

<fantasai> ::distributed() matches pattern of ::cue() and ::region(), seems we're alinging on that

<fantasai> ScribeNick: fantasai

dg: @host did not solve the body class=light, and have components be able to see that

sjmiles: ANd we never used this @host { * {}}

fantasai: You probably want ::shadow, ::light, and ::context (to reach out)

dbaron: or a combinator to jump out

fantasai: Issue with a combinator is that it breaks the rule where combinators limit the matched set as you go
... so if you did bar <magic cobminator> foo, suddenly you're selecting a different set of foos

dbaron: What I was thinking of was a combinator that would let you get to the scoped root from the selecto rthat's selecting inside it
... which is adding restrictions, right?
... in the dark theme use case

tabatkins: yeah, that works with combinator
... eventually we rejected that idea

[tab opens a Google Doc to show this idea off. He will share a link here]

[we will share the doc later]


scribe: does that work?
... no :-(

tabatkins: host element and shadow have equal claim
... we need to pretend the hos telement is in the root of the shadow tree
... so i fyou want to select on it, all you do is [writes in doc]
... this example will target the host element outside


^ that link works


tabatkins: you want to be able to select based on the content further up in the document. like the theme use case, or modernizer up higher
... but you don't want to allow arbitrary selecors above

<jkomoros> can you see this?

<dbaron> ScribeNick: jkomoros

<singhalpriyank> yes

tabatkins: If you have the outer document followed by a shadow element, and inside of that another component
... the outer document you would see includes the shadow tree of the outer component.
... that breaks encapsulation, allows developers to depend on details of components outside
... we still need a communication channel to outside
... we think we have a simple thing that satisfies use csaes
... here's an example. The context pseudo class is placed on the root element (hsot element)
... it matches if something in the compound selector matches in the fully composed ancestor list (?)
... including stuff in other boundaries
... because maybe you're applying a theme inside of one of the parent components above
... it allows some information to be piped through, but not enough to allow a fragile dependency (we hope)
... the list of elements checked starts with host element itself, goes up to the root, through any of the composed shadow trees
... that's the only way to select up outside
... going the other way, we still use ::distributed
... works the same way
... we think this solves all the use cases we know of
... and it's convenient and easy ,not the contorted tree hopping of @host and everything else

dbaron: What is removed?

tabatkins: What is removed is the @host (in favor of moving host element into shadow tree for styling purposes, and using context pseudo class to select up)

rniwa: So if you have multiple composed layers, it selects each composited layer (?)

tabatkins: no, the fully flattened tree up above
... you do see the shadow dom of things up the tree... but not very much
... so at any point you can inject information in

rniwa: So a shadow DOM A, inside sahdow dom B

tabatkins: we haven't changed the way normal selectors work. In a shadow style sheet, still only match within that shadow tree
... only thing that I think we might want to change, as a result of the recent discussion aroudn region pseudo-element (as opposed to rule)
... the problem is this distributed pseudo class isn't compatible with any nesting mechanisms we might add in future
... like, if you had foo bar baz {} as foo { @nest bar baz { ... }} , the distributed pseudo class wouldn't let you do the nested selectors
... same problem applies to regions, because regions often have complex selectors inside of the regions
... a possible aternate syntax is to have content selected and inside have a @distributed rule that takes ...
... content { @distributed { :scope > foo {}}}
... behaves similarly, but more future-compatible
... it's an @-rule inside of a declaration block
... we agreed to use it in error handling rules. This would be the first other use of it

es: I like the ::distributed

sjmiles: Yeah, that one is easier to type

fantasai: what about ::distributed <space> <other stuff>?

tabatkins: problem about jumping sub-trees, not narrowing matching
... if that's not a problem, then maybe that's fine

es: That space one requires deeper architectural change to selector matching

sorvell: I don't think the notion of limiting across selectors is something web developers know or care about

dbaron: I don't think it violates it, although ::distributed is the wrong name in this formulation

fantasai: light?

sjmiles: well, one person's light is another person's shade

es: As an implementor I don;t like it

dg: We have the same basic thing with pseudo elements already
... we take this linked list and grab and swap it around at the end

<fantasai> fantasai: It's just a syntactic difference; implementation can store it in whatever structures it wants

es: yeah, but this would come at the end

es; Why is this not an @ rule?

scribe: like @teleport

<dbaron> I'd rather have content::back-to-light-dom > .foo { ... }

tabatkins: You don't want you to accidentally select hidden things in shadows above you

dbaron: I think that we want selectors that continue to the right of pseudo elements. And the implementation model there is treat them like you treat pseudoelements today, where you match the thing to the left first, and then you
... say , oh, pseudo element, do this other stuff
... I think it makes sense without parens

es: But selector matchign starts from right side

dbaron: not really, not with pseudo-elements. You have to start from just to the left of hte pseduo-element

<stearns> the key is that we're combining two selectors. You can still use right-to-left evaluation on each piece

dbaron: now we're going to allow more stuff to the right of ::, but still same model

es: why is the other thing not good as a functional syntax but this is?

dbaron: Because it's a singular thing (?)

es: The current distributed thing matches cue

dbaron: Tab doesnt want's functional syntax because nested syntax will come along in the ftuure

es: I'm not comfortable with rewriting whole selector checker

dg: You just do it when parsing rules

tabatkins: Find first pseudo element, run part before it, then ... [didn't get]

es: But it's not "at end", it's a nesting relationship. It's more complicated

tabatkins: exactly like a b is not all b's just b's inside of a's

dbaron: I agree it's hard. I think we want implementation experience on concept before we commit to using it
... but it's the same concept we've come up with in multiple places already (like overflow fragments, here, regions)

fantasai: cue?

es: cue currently works like distributed does
... distributed is consistent with that
... the inner selector in there is not even HTML, it's a totally different world

tabatkins: But different constraints: the document exposed is completely flat
... whereas this will expose more complex things inside the parens

<dbaron> It needs to be called the see-you-eee element (the "cue" element) and not the queue element (the "q" element)

dfreedm: I've hit this before. Nesting would be great

es: What people are arguing for is a "reuse this selector" ability in CSS, like a #define for selectors

dbaron: But with that, you'd end up having something with an unmatched parens in your #define,

tabatkins: yeah, that would be painful

sjmiles: Looking at the multiple {} solution, if I write that rule, I might be tempted to ask, can I put stuff to the right that is different than what's to the left? (?)
... as a developer, it's just getting in my way. Confusing.

es: In this syntax, how do I match stuff that is a sibling of the stuff that's distributed
... example: content::distributed(> .foo) + span {}
... what does that do?

tabatkins: That doesn't do anything in current syntax
... given the assumptions of functional syntax, we're doing that because "only one pseudo element, and at end rule". So this is nonsense, because it comes after pseudo-element
... but content::distributed > .foo {} is also nonsensical

dbaron: So you want a pseud-class instead of a pseudo-element?

es: I want to style the heading element that immediately follows the first heading element

tabatkins; The general use case of dropping down to jump back up (?) is a generic discussion not limited to this discussion

es: I'll reserve judgement, but I don't know what happens if you have two distributed

tabatkins: You can never ahve a double distributed, because content nodes are gone

dg: yes you can

<fantasai> content:matches(!::distributed > .foo) + span

dg: imagine that you're inside of a tree that's inside of a shadow tree

tabatkins: But as far as you can tell, you're not in a distributed tree (?)
... content::distirbuted > .foo::region p content {} will never match anything

es: ... no?

tabatkins: remembe,r the :context selects on flattened tree
... below you, any contents you contain you can't access content

es: That's not how it's currently specced. It's currentlys pecced that elements are distirbuted, but not that <content> is gone

tabatkins: But it doesn' matter for the purpose of this example

dg: What he's proposing works the same way as one with parens, just no parens

es: So in <an example> you could have interleaved with multiple implied parens

dg: Positive impression from CSS people around dropping parens?
... what about people who implement?

dbaron: It doesn't seem easy, I think we should get implementation experience before we commit, but I think it's probably the right thing

tabatkins; So we leave spec as it is right now, we add notes with paren-less version, that says implement and give feedback, if it does work then we use it

scribe: someone has to solve these similar problems (e.g. in regions) first that leads the solution

<stearns> I'm happy to change to this

es: So you change region, and hixie changes cue?

dbaron: cue might be a special case? it's selecting into a different document

es: I want to hear from apple

rniwa: We don't like ANY changes. ideally we wouldn't implement anything, but we'll have to implement SOMETHING

dg: I'm interested in how hayato-san feels about this
... and see how much he screams

eo: my rule of thumb is to let dbaron do his experiment and see what happens

<scribe> ACTION: tabatkins to update the spec to the paren-less version of the :distributed, with a note that we will use that syntax if implementors don't scream after experimenting with implementation [recorded in http://www.w3.org/2013/06/21-webapps-minutes.html#action01]

<trackbot> Error finding 'tabatkins'. You can review and register nicknames at <http://www.w3.org/2008/webapps/track/users>.

es: It's defiintely implementable, it's a question if the implementation cost justifies the developer confusion benefit

dbaron: Remember, either of these solutions is hard

dg: so instead of a list, it's a tree
... so what will happen is that at parsing you'll have to be aware that when you see this pseudo-element you change what you've seen already into a tree and parse selector again

es: One of these was described in a grammar. But the current proposal can't be done in a grammar; it's context sensitive
... so it makes it harder to implement. The parser has to be made more complex

(in the action above)

dg: context is confusing, because it looks like content

es: what about "projection"

rniwa: THat's too complicated

dbaron: Let's get rid of context entirely

es: what about ":path"

dg: ":composed"

dfreedm: I prefer path

es: "has" looks closer to "matches"

sjmiles: Front end developers want everything to be as short as possible

rniwa: What about ":host" since we got rid of @host?

sjmiles: not bad

dbaron: agreed

es: But this could be arbitrary levels above

rniwa: I like ancestor

eo/sjmiles: I find it confusion

fantasai: I like host best

es: What about :inside?

sjmiles: that's the opposite of how we think about it as devleopers

rniwa: yeah, I'd expect that to be OUTSIDE

es: If we do distributed shenanigans, why don't we do same thing here?

dbaron: This is intentionally limited

<dfreedm> too late

dbaron: I think we're moving to :host?

tabatkins: not bad

es: but x is the host here, and the theme is on body
... path makes sense, like a traversal path

fantasai: What if you allowed host element to be matched in host

tabatkins: you can: :host(*)

sjmile: Is there a way to avoid me having to write "x" all the time for my placehodler
... now we don't use the name, we just use @host

tabatkins: no need to worry about it. will only match pseudo element

es: is there a way to reference your host without explicit tag name?

tabatkins: you can: :host()

es: how does that differ from :scope

tabatkins: no, because things aren't actually scoped
... the shadow is not actually a scoped style sheet
... it happens to be scoped, but it isn't technically a scoped style

es: I think we should go with :host()

tabatkins: We can omit empty parens

rniwa: I like :host

proposed resolution: :host(<simple selector on ancestor path>), or :host, which is equivalent to :host(*)

fantasai: What's the specificity

tabatkins: I think we can add the specificity inside the host (?)

<dbaron> I think s/simple selector/chain of simple selectors without combinators/

es: what about like [data-foo]:host(.dark)

CONCLUSION: :host(<chain of simple selectors without combinators on ancestor path>), or :host, which is equivalent to :host(*)

sjmiles: This is a different concept than ancestor. It has some similarity to "something that's above me"
... it feels a bit weird to put what would be on the left side would be in the parens to the RIGHT

whoops, sorry strike that resolution

I misunderstood what "resolution" meant in this context

thanks for the information! I'll get the hang of this some day :-)

sjmiles: To be clear, this syntax is fine, but ultimately developers would have wanted somethign similar: .dark goes on left, then some host, then wormhole
... but this is fine, given all the constraints.

<dbaron> just wait until I propose :上(.dark)

tabatkins: earlier we had a ^ combinator which said (jump boundary), but it was weird because you could have only one simple thing on the left

<dbaron> or :下(.dark)? Not sure which makes more sense.

tabatkins: I think it's easier to internalize the restrictions that things inside the parens play by different rules than the things on the left of that magic combinator

rniwa: I agree the ^ is weirder than :host

[discussion has gotten disorganized; scribe has been unable to keep up for the past 2 minutes]

fantasai: We should agree that words should either be plural or not plural

<fantasai> fantasai: And CSS already has 'content' in some places

CONCLUSION: rename ::distributed to ::content

sjmiles: you don't need to have the content ahead of the ::content

fantasai: This helps people remember what it means, what concepts it's connected to

<fantasai> (because the tag name is content -- distributed is just out of the blue)

rniwa: Ideally the developer needs to know the minimum of terrms

<dbaron> so 'content: contents' in http://dev.w3.org/csswg/css-content/#contents0 should be 'content: content' ?

<fantasai> either that or something else, yeah

[lunch break]

[lunch break over]

dg: Any more CSS - related topics?

dbaron: I don't know if we agreed if custom pseudos should be functional or not?

eo: Yes

es: I think it would be said to sacrifice explanatory effect

dg: I agree with that

eo: I think it makes sense; it calls out that pseudos you come across are from a component somewhere, as opposed to from the system
... there's a clarity that gets sacrificed by confusing the two

tabatkins: Even if we use no (), we still need to have a prefix (?)
... even the explanatory power of a non-functional syntax is still limited because custom ones would have to be named in a way that wouldn't conflict with new ones

es: Okay, I won't fight for it

dg: dbaron mentioned a solution I'm okay with
... have a switch in shadow DOM spec that says that UA can define pseudos if they want

es: Long ago, the idea that in this future world all the quirks of the bedrock go away, evertyhing is components and who cares
... but we're leaving warts all over the place

sjmiles: There's a tension. Sometimes it's a clarity concept, but sometimes user doesn' tneed to know

eo: In the future with lots of components, you can clearly tell when you're dealing with a part of a components

es: Why need a prefix? Why can't you implement ::placeholder

tabatkins: We want to avoid namespaces that are unchecked, so that we don't have to worry about compat checks for new keywords in the future

es: Why not expose things that are actually inside of shadow roots, why aren't they just a namespace open to anyone?
... placeholder is specced
... you'd need to say ::part(placeholder) (?)

eo: you wouldn't want to expose it as a shadow root; it's an implementation detail

es: But every browser implements placeholder in the same way
... and this argument is, what if someone hypothetical browser wants to do it some other way?

dbaron: Placeholder in gecko uses native anonymous content
... that's different from XBL, which is different from web components

es: Gecko creates divs inside of an input element, right?

dbaron: The inside of a select box is more complicated. There are things in there that have boxes but no content nodes for those boxes
... which is different from Native Anonymous Content
... because NAC is where we construct actual content nodes and construct boxes for those
... for a select we just make boxes

es: I don't understand the harm in claiming that <input> uses component model
... like, keep ::placeholder. But why not also allow the new syntax to address the same thing?

eo: Because the implementors can choose how to implement it!

dg: The question is not, whether it's implemented with Shadow DOM, but rather how it interacts with rest of content
... so you don't have all of these weird behaviors and edge cases. It's all described by semantics of shadow DOM

dbaron: You assume shadow dom will be web compatible

dg: We have implemented almost all crazy elements uses Shadow DOM machinery
... everything exceltp select, which we have plans to switch over as well
... it's been for two years now that we've been using it

es: If I have input::part(placeholder) and it was a normal input, now I inherit it, I shouldn't have to change the CSS that targets it; I should just have a placeholder pseudo in my new shadow dOM

[murmurs of aggremeent from sjmiles, dg]

dg: yeah, big fan
... part() comes from an old discussion

dbaron: There are three sorts: pseudos that are not tree like, some that are leaf like, some are tree-like and contain real elements
... CSS doesn't define any in the third category. But there's been discussion about various ones

es: ::first-line can't possibly be implemented as an element, for example
... it's different from somethign that can be exposed as an API surface to style

dfreedman: I don't understand why it has to be different

eo: part() takes an author-defined ident

sjmiles: I think it's a strong point: this is a specific kind of styling we're doing here--it's a node. It's the same for web components and implemented pieces
... it seems this makes sense. It makes sense that ::first-line is different

dg: it's hard to explain ::backdrop in terms of these primitives
... so that would stay the same

es: like there are some things that are clearly defined by what's inside
... like scrollbars

eo: I disagree about scrollbars

dbaron: We plan to get rid of all internals to scrollbars

es: What kinds of scrollbars would you not want to build in the div

rniwa: That's for the scrollbars that have been released to date; in the future they might not

eo: Currently in WebKit if you use legacy pseudos for scrollbars, you go to legacy scrollbar mode. If you don't, then you get the special new ones

es: What I'm saying is that if you don't want the API surface, don't expose it. In this case, it's already been exposed

sjmiles: I think it's reasonable to say, for this thing, the way we implement it we shouldn't reveal it

eo: I want to make a distinction between UA-implemented things, and things built in WC world. We want that distinction

es: But why? If the standard says ::placeholder, it cannot be removed entirely
... so for new ones, it should use part() so that others can override later. But if it's not exposed, doesn't matter

dg: it gives a clear message to author of how to reason about it--it's just like Shadow DOM.
... the UA is acting as an author in those cases where the guts are exposed like this

rniwa: We don't want to be adding APIs that limit the ways in which we can tweak UIs

dg: agreed. I don't think it matters. ::placeholder behaves like it's implemented as Shadow DOM. So it should behave that way fully

eo: The entire existing platform prior to the feature means that user-land psueods and UA pseudos are different

es: today in WebKit, ::-webkit-spin-button IS a div today. That's an API surface
... right now it has a prefix
... but if a spec adds a non-prefixed api surface, it's now a real API surface and it should behave like the other non-magic things

eo: as an author, I want to be able to send a bug report to the right person

dbaron: Authors aren't the users. End users are the user. Web platforms are designed for both of those constiunenciies. Sometimes you don't want to give control

es: But we're talking about, for things that ARE already exposed

dbaron: Some of the motivation for it, is there's stuff that might be the same now, but we don't want to be stuck with it forever

eo: we're stuck with legacy scrollbar contrrols forever. :-(

dbaron: mobile use cases are a great example. I'm glad we didn't commit to select's having a dropdown button
... I think you're being conservative about what to expose

es: You seem to be arguing that ::placeholder was a mistake and we should remove immediately?

dbaron: I think freezing everything about form controls at this moment is a bad thing.

dg: I think we're talking about two different things

sjmiles: UAs should be able to choose what to expose. But, IF you have chosen what to expose, like ::placeholder, then what you expose is Shadow DOM

dg: ::placeholder already behaves, as specced today, as a custom pseudo-element
... as if it was a custom pseudo-element

rniwa: Spec doesn't say that ::placeholder needs to be text inside a Shadow DOM. Just that color and whatever can be controlled
... this allows UAs to match the convention of the platform
... placeholder is okay, but in case of like scrollbars, we have to show legacy scroll bars even on platforms that have other scrollbars --it's confusing ot users

es: I don't care so much about scroll bars; mozilla doesn't want to implement them
... I'm saying when it's a standardized API surface, it should be exposed with Shadow DOM

eo: take input type=password
... users before authors, etc
... it would be reasonable for an impelemnter to say that user type=password cannot be overriden by components

es: that's fine, the standard should say that for those cases

dbaron: I think there's another answer to this requirement
... your requirement is about re-impelmenting existing parts of the platform
... I'm okay with the component model having a way to match an existing part of the web platform
... that's different from saying that you can extend the pseudo-element space arbitrarily

es: To clarify, I'm arguing we should do ::part(). And that we should fix the platform to fit this

eo: Dbaron is saying that I want to impelement a new component and react to ::placeholder. Because it's an existing platform feature, you should have a way to say, this thing is my ::placeholder
... and then if you add a new pseudo, it goes in the box where user-land new pseudos go

sjmiles: Eliot's proposal does everything yours says, but simpler, from our perspective

es: We all seem to be talking from different contexts

sjmiles: I think we're closer than we think we are

dbaron: eliot wants ::part(placeholder) to react to all the same thigns that ::placeholder does. I think they should be two different name spadces, and Shadow DOM should be able to get to both of them

sjmiles: But you're telling Bob Web Developer that in case A you use this syntax, but in case B use a different one. But there's no difference
... why does it matter what comes from UA and what comes from a component?

rniwa: That is an API surface. It's a namespace issue

es: It doesn't matter if you don't reproject

dbaron: the reason I don't like them being the same, that goes back to the assumption you'd implement all of the platform in Shadow DOM

sjmiles: I think rniwa has a good point about namespaces


rniwa: We could have collisions with new formalized pseudos if they share the same namespace

es: ::part(placeholder) is exactly like a method that gets stomped on by some new web platform feature adding a new method to a type of node
... those types of collisions happen all the time.
... You're trying to protect yourself from one, particular type of namespace collision. THere are lots of them!

dbaron: There's a different error-handling rule for pseudos, and WebKit implemented something else, and didn't come back to the standards groups
... now you come back and say you want to build a new feature on top of that broken feature

es: We agreed on that part--we should move to ::part(placeholder)
... isn't this like saying, custom elements are bad idea

eo: yes! I think you should only do it if it's a good reason

dfreedman: but how do you know in advance if they'll have a good reason?

es: I'm saying, all platform widgets, however they're implemented (jquery, etc) they should have the same API surface
... not if you use Shadow DOM surface.

dfreedman: part(oplaceholder) just describes the same thing as ::placeholder

dg: eo is saying htat part defines user-defined pseudos

dbaron: I think I'm okay with either formulation (?)

rniwa: think about for example, translation attribute
... we shipped and broke a lot of libraries in india

eo: We have compat constraints we need to take into account

rniwa: We want to avoid breaking existing websites

es: If we go there, and say part is magical only userland, and every pseudo in user space should start with data- so you don't conflict
... so you're saying that every method in say polymer has to be polymer_foo()

sjmiles: We tried somethign at the beginning, where we presented a different public face. Developer users HATED it
... developers want more power AND backwards compatibility. It's hard.

rniwa: MediaWiki added a special style attribute to work around WebKit bug
... we fixed it, but now we can't remove it because mediawiki content is deployed with the old stuff

eo: Quickly: this is the same argument from last F2F: the idea if we should allow custom element names over the wire
... is=foo nicely delineates which is user land

dg: it's solved with the - requirement

rniwa: I'm less concerned if authors can only extend HTMLElement

es: Either developers are responsible for avoiding namespace collisions everywhere, or nowhere

sjmiles: Seems like everyone agrees the ::part() is a good idea in many cases

CONCLUSION: ::part() is a good solution for user-land pseudo elements
... although the group doesn't necessarily agree about existing pseudo elements at this point

dg: ::Part and ::pseudo is fine (?)

<dbaron> (by either formulation -- one formulation is that ::pseudo and ::part(foo) are just two different namespaces, shadow DOM can't extend the first namespace, but it can create things that match either -- the other formulation is that ::part(foo) is an extensible namespace and future Web features that could be represented as such a part (whether or not they are implemented in terms of shadow DOM) plus ::placeholder should have such parts done as ::part(foo), an

<dbaron> d that ::part(placeholder) should be an alias for ::placeholder for consistency

<dbaron> )

CONCLUSION: the attribute to register these custom pseudos is "part" (not pseudo as it currently is in spec)

The CSS portion of this discussion is over. We'll stop taking notes h ere

<MikeSmith> .win 15

Summary of Action Items

[NEW] ACTION: tabatkins to update the spec to the paren-less version of the :distributed, with a note that we will use that syntax if implementors don't scream after experimenting with implementation [recorded in http://www.w3.org/2013/06/21-webapps-minutes.html#action01]
[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.138 (CVS log)
$Date: 2013-06-25 11:17:50 $