[css-cascade-5] Cascade Layers need CSSOM support

[anttijk]

CSSLayerRule interface needs to be specified.

[xiaochengh]

This is not entirely trivial. Since we have two types of @layer rules, the empty statement and the block rule, do we want the same interface for both of them, or different interfaces?

I don't have a strong preference. Just putting the options here for discussion.

---

If we use different interfaces, the following seems pretty straightforward:

[Exposed=Window]
interface CSSLayerStatementRule : CSSRule {
  readonly attribute CSSOMString names;
};

[Exposed=Window]
interface CSSLayerBlockRule : CSSGroupingRule {
  readonly attribute CSSOMString name; // empty string for anonymous layer
};

---

If we use the same interface, it would probably be like

[Exposed=Window]
interface CSSLayerRule : CSSGroupingRule {
  readonly attribute CSSOMString name;
  readonly attribute boolean isEmptyStatement;

  unsigned long insertRule(CSSOMString rule, optional unsigned long index = 0); // Is this the correct syntax for override?
};

We need to override CSSGroupingRule.insertRule() so that it throws an exception when called on empty statements.

[tabatkins]

I think we should make them separate objects, since they have quite distinct data models - one vs N names, child rules vs not. They're the same name for authoring convenience, but that doesn't need to translate to them being the same object type.

[emilio]

I'd prefer not having two interfaces for this, fwiw... Also, for the @layer a, b, c; case we need a sequence of names, not just one name, right?

[tabatkins]

Yup, that's what I meant by "one vs N names". @xiaochengh's example supports that, it just exposes the N names as a single CSSOMString rather than as a list of them.

[anttijk]

Having two separate interfaces might actually make sense as there is nothing that can be shared.

CSSImportRule also needs an extension.

[emilio]

Yeah, I guess that's fine / alright, specially given we've given up on making CSSRule.type return something meaningful for these.

[anttijk]

On the other hand these interfaces are pretty useless anyway so I'm equally fine with a generic object that exposes nothing but cssText and the tree structure.

[emilio]

That'd be totally ok for me too.

[emilio]

FYI. Not ideal, but I needed some OM support to make DevTools etc work in Firefox.

What I did for now is implementing it as:

interface CSSLayerRule : CSSGroupingRule {};

Where cssRules will return null for statement layers. Again, probably not ideal, it was just the easiest to implement.

[anttijk]

interface CSSLayerRule : CSSGroupingRule {};

Did the same thing in WebKit.

[xiaochengh]

Can we prioritize this issue? We need a spec to move forward.

[mirisuzanne]

Added agenda+ to get a resolution on the approach.

[css-meeting-bot]

The CSS Working Group just discussed Cascade Layers need CSSOM support, and agreed to the following:

• RESOLVED: Prop: We will have 2 separate objects; one for rules w/o a block and one for rules with a block

The full IRC log of that discussion

Topic: Cascade Layers need CSSOM support
github: https://github.com//issues/6576
miriam: A few options shown in the first ocmment on the issue. Discussion about 2 objects or 1 single. If 2 one is for empty layer rules and the other for layer rules with something
miriam: Don't have an opinion, if anyone does love for them to speak up
astearns: Looks like people in issue are saying whatever
miriam: Seems no strong preference. A little sense of it's not useful but needed
astearns: That 2 objects are needed?
miriam: That some resolution is needed. But not particularly useful either way
q+
fremy: Maybe idea for simlied. @layer a with no rule and @ layer b with no rule and you don't need one or the other. Then it's only one type of rule. I don't write these, but maybe an idea?
ack heycam
q?
heycam: With a single interface...perhaps issue with both approachs...a bit weird if insert and remove rule can change the type. Maybe argument for one interface
was suggesting maybe considering `@layer a, b, c` syntactic sugar for `@layer a{} @layer b{} @layer c{}` then you only have one type of rule
heycam: I see suggestion from xiao was throw exception when insert called on rule without block. I guess need inverse as well
note that @layer b {} is not valid before @import, but @layer b; is
astearns: Have not thought much but slightly more in favor of single interface that could expect both and lets you go back and forth between
heycam: Reasonable for author to say start with a layer that has no block and then want to, in cssom, add children?
@fantasai: ah, ok, then that doesn't work. except if we change it to allow @layer {} before import but not allow it to have rules in it?
heycam: If we want to allow that I guess the single interface makes more sense. Else no way to preserve object identity. Perhaps that's not reasonable
ack fantasai
q+ to talk about syntax mismatch
fantasai: The @layer rule with a block that accepts style ruels is not allowed before @import but is allowed after
fantasai: There's several different syntax restrictions
fantasai: I do think heycam suggestion makes sense. Likely authors will want to add and remove rules from @layer. If converting from being a block ro not I'm not sure, but adding to an empty block I can imagine
fantasai: There's complication there about being before or after @import
heycam: Symantically important to allow @layer block before @import?
fantasai: Yes b/c layered in order of apperance. If importing rules that define appearance it matters. Layers with same name combine.
fantasai: If you want rules before the @import you can add the layer and have it appear before @import
ack TabAtkins
TabAtkins, you wanted to talk about syntax mismatch
TabAtkins: fantasai covered some of this, but more syntactic mismatch. @layer that doesn't have block allows comma sep. list but with block is only 1 name.
TabAtkins: Closely related but syntax is difference. If allowing to modify they need to be 2 objects
astearns: For blockless do we need object to represent that or multiple blockless each with one name
TabAtkins: Don't have precedent for single rule expanding to multiple. Not impossible, but prefer avoid
astearns: Sounds to me like need 2 objects, one for blockless and one for block
astearns: Other comments on just that? Then figure out mutability and errors. Single or 2 objects. Anyone continuing to argue for single?
astearns: Prop: We will have 2 separate objects; one for rules w/o a block and one for rules with a block
astearns: Obj?
RESOLVED: Prop: We will have 2 separate objects; one for rules w/o a block and one for rules with a block
astearns: Object with a block we should be able to add. For blockless should we add an error or should it not have methods?
TabAtkins: Won't have methods. No child or insert rule
astearns: miriam anything else on this? Or should we go with resolution and see what we need on future calls
miriam: Nothing else

[fantasai]

It seems pretty clear that we'd have (as @xiaochengh outlined above):

[Exposed=Window]
interface CSSLayerBlockRule : CSSGroupingRule {
  attribute somethingsomethingforthenames;
};

[Exposed=Window]
interface CSSLayerStatementRule : CSSRule {
  attribute somethingsomethingforthenames;
};

But the open question is, what are the somethingsomethingforthenames?

Tab suggests to have .name or .nameText or something on the group rule be a simple string (e.g. "name.subname"), which is pretty straightforward, but then for the statement rule, do we want a single string ("name1, name.subname, name3") or a list of strings ("name1", "name.subname", "name3")?

It feels pretty weird if we have the attribute on these rules both be single strings while the group rule string has commas in it and the other one doesn't. Miriam also thinks it feels more natural to have an array here. We'd prefer something like string .name for the grouping rule, and an array of strings .nameList for the statement rule. (But neither of us knows much about CSSOM interfaces.)

One point that @tabatkins brought up was that various comma-separated things in the CSSOM (e.g. selectors, conditionText) are represented as a single list. But a) in these cases the comma represents an "or" operator, so the string is not so much representing a list of things as a single combined condition and b) none of these are trying to coordinate with a nearly-identical rule whose corresponding attribute represents a single item of the type represented in that comma-separated list.

[xiaochengh]

We'd prefer something like string .name for the grouping rule, and an array of strings .nameList for the statement rule.

I think this is better than what I previously proposed. I wasn't aware of the array type in IDL.

[anttijk]

We'd prefer something like string .name for the grouping rule, and an array of strings .nameList for the statement rule.

I like these names and return types. They also match the names WebKit uses internally already.

[emilio]

Sounds good. Should the attributes be readOnly as per #6645? It's not a massive deal to support mutable attributes, but it's almost always a footgun.

[anttijk]

Yeah, should be readonly.

No new mutable CSSOM should be added ever, really. It is a useless footgun.

[tabatkins]

Ah, if we're doing readonly, then my objections to arrays mostly go away. Having mutable arrays requires a lot more logic that isn't really justified here. So we can just do a readonly FrozenArray for the type on the CSSLayerStatementRule and be good.

I still suspect having the layer names themselves be a single string with the possibly-dot-separated name in it is best here; most of the time you want to treat the name as a unit anyway, and when you do want to break it down, a simple name.split(".") is reliably correct unless someone put an escaped period in their layer name like foo\.bar (which, uh, stop hitting yourself).

[css-meeting-bot]

The CSS Working Group just discussed [css-cascade-5] Cascade Layers need CSSOM support, and agreed to the following:

• RESOLVED: add the cssom rules as .name and .namelist

The full IRC log of that discussion

Topic: [css-cascade-5] Cascade Layers need CSSOM support
github: https://github.com//issues/6576
https://github.com//issues/6576#issuecomment-943858864
miriam: My understanding is we wanted 2 interfaces, one for statement and one for block rules
miriam: block rule has read only attribute of the name of the layer. statement would have read only frozen array of strings
miriam: Does that cover it?
q
fantasai: Question was what to call them. .name or .nametext vs .namelist
TabAtkins: .names
fantasai: .name and .names seems tricky
TabAtkins: yeah
fantasai: .namelist regardless of the other being .name or .nametext
Rossen_: .name is shorter
fantasai: great. resolve on that
Rossen_: obj or suggestions?
RESOLVED: add the cssom rules as .name and .namelist

[mirisuzanne]

@anttijk @xiaochengh @emilio let us know if the edits in 6e8482b satisfy the question here.

[foolip]

For ease of review, that's https://drafts.csswg.org/css-cascade-5/#layer-apis.

[xiaochengh]

b7c7a15 lgtm

[xiaochengh]

Wait, there's one last thing missing: We need to be able to distinguish whether a CSSImportRule is unlayered or in an anonymous layer.

How about making layerName nullable (readonly attribute CSSOMString? layerName;), so that it returns null when the rule is not layered?