[css-syntax-3] "Consume a block's contents" leaves decls list with

[AtkinsSJ]

Consume a block's contents collects declarations into the decls list, but only copies that over into rules when an at-rule is encountered. If decls is not empty when <EOF-token> or <}-token> is encountered, those declarations are lost.

I think the fix is to change this:

<EOF-token>
<}-token>
Return rules.

into this:

<EOF-token>
<}-token>
If decls is not empty, append it to rules.
Return rules.

This at least worked for me.

---

As a side note, not sure if #8834 is asking for this particular kind of feedback, but the new algorithms have been very straightforward to implement directly, and feel like a nice improvement to how they worked previously. (Token streams are how I'd ended up implementing it before so it's nice to have that actually match the spec!) So thanks @tabatkins

[cdoublev]

Considering that declarations at the top of a style rule should not be wrapped in a nested declarations rule, I think this algorithm should either still return a list of declarations and rules, or CSS Nesting should specify that this top-level list should not be included in CSSStyleRule.cssRules but exposed as CSSStyleRule.style.

[cdoublev]

Actually, many other rules only have declarations, so it should definitely return rules and declarations, imo.

[AtkinsSJ]

I can't remember if that was confusing to implement or not. I just ended up treating a list-of-declarations at the start in a different way. But being more explicit is always nice.

[cdoublev]

I am still unsure and considering all the options.

On one hand, a representation like Contents(Declarations) as the result from parsing against <declaration-list> seem quite odd. On the other hand, all rule block might be considered as containing rules, and lists of declarations implicitly wrapped in an invisible rule. But then it is not clear why the top-level list might not be included in the list of rules...

[AtkinsSJ]

Let me know if you have anything you'd like to test and I'll give it a go. Ladybird tries to implement specs as closely as possible, so it should be a pretty good test of whether the spec algorithms work or not.

[cdoublev]

I am in a similar position. I was just thinking out loud about what the appropriate spec text should be. I am not in a position to decide it. 😸

[AtkinsSJ]

Oh neat! I hadn't noticed sorry. 😅 I might have a look at that for some ideas.