[css-2025][css-conditional-4] Add CSS Conditional 4 to Reliable CRs

[SebastianZ]

Splitting the discussion from #9770 for adding CSS Conditional 4 to the Reliable Candidate Recommendations section of the CSS Snapshot 2025.

Data to consider:

Current status: CR (last published: 17 February 2022), ED from 31 August 2025
WPT: passing tests (stable): Chrome: 3041, Edge: 3109, Firefox: 2671, Safari: 2839; interop: 2622 / 3202 (81.9%)
Current Snapshot section: Rough Interoperability
Open issues: on GitHub: 4, in spec.: 1

Summary: High interoperability, spec. in CR, few open issues

Notes: This spec. is currently still a delta spec. against Level 3. And Level 3 is currently part of the Official Definition.

Sebastian

[svgeesus]

WPT: passing tests (stable): Chrome: 3041, Edge: 3109, Firefox: 2671, Safari: 2839; interop: 2622 / 3202 (81.9%)

The ED of Conditional 4 shows 31 tests. Although Bikeshed says

There are 272 WPT tests underneath your path prefix 'css/css-conditional/' that aren't in your document

and some of those might be conditional-4 tests, too. I will go through them to investigate.

Edit: Most are Conditional 5 tests for container queries.

[svgeesus]

The ED of Conditional 4 now shows 41 tests.

Chrome 139 40/41
Firefox 142 34/41
Safari 18 35/41
Edge 139 40/41

[svgeesus]

This should be republished as CRD, because there is one useful clarification and also the test coverage is better than the 2022 CR Snapshot.

[SebastianZ]

WPT: passing tests (stable): Chrome: 3041, Edge: 3109, Firefox: 2671, Safari: 2839; interop: 2622 / 3202 (81.9%)

The ED of Conditional 4 shows 31 tests. Although Bikeshed says

There are 272 WPT tests underneath your path prefix 'css/css-conditional/' that aren't in your document

The ED of Conditional 4 now shows 41 tests.

Chrome 139 40/41
Firefox 142 34/41
Safari 18 35/41
Edge 139 40/41

Thank you for pointing that out and updating the list of tests!

Disclaimer: The numbers I am referring to in my stats in the different issues regarding CSS 2025 relate to the overall coverage of the related Module, mostly not restricted to the features of a certain level.
Given the numbers above and given, the passing tests are overlapping, we currently have a slightly better interop rate of 82.9%.

Sebastian

[svgeesus]

An updated CRD should be published Thursday 4 September.

[css-meeting-bot]

The CSS Working Group just discussed [css-2025][css-conditional-4] Add CSS Conditional 4 to Reliable CRs, and agreed to the following:

• RESOLVED: css-conditional-for to Reliable CR section

The full IRC log of that discussion

<ChrisL> q+
<emeyer> SebastianZ: this doesn’t have a very high interop (82%) but only has a few open issues on the spec
<emeyer> …question is, what’s the threshold to moving to reliable CR
<emeyer> ChrisL: We should look at the spec and tests; thaere are only 41 tests
<astearns> ack ChrisL
<astearns> ack fantasai
<ChrisL> s/the spec/only this level of the spec/
<emeyer> fantasai: An original goal for Reliable CR is to distinguish “this is almost a rec but we didn’t get close enough”
<emeyer> astearns: It is a good question to ask what the interop threshold should be — 80%, 90%, a particular number of tests?
<ChrisL> 35 × 100 ÷ 41 = 85.3658536585
<emeyer> fantasai: I don’t think percentages work, because sometimes they’re mostly parsing tests and may not fail when the spec isn’t supported and don’t say what’s actually failing
<florian> q+
<emeyer> …what you’re looking for is, “Is this really close to being a recommendation but there are a few remaining bugs or well-understood minor issues?”
<emeyer> …Has it been getting enough attention that bugs are likely to be found? Or is it a mostly-ignored spec?
<ChrisL> q+
<emeyer> …When you have a lot of passing tests AND you know the feature is well-implemented across at least two implementations, that’s good
<SebastianZ> q+
<emeyer> …Just looking at test numbers may not tell you what you need
<emeyer> …Layers is one where I’d expect good numbers to be a high signal, but for layout you may have a lot of processing tests but not many rendering tests
<astearns> ack florian
<emeyer> florian: I think you may have mixed up categories a bit
<fantasai> or things like media queries or ui stuff, where you're more likely to have extensive processing tests than behavior tests
<emeyer> …I think we tend to be fuzzy about which level is for what, which isn’t a great thing
<emeyer> fantasai: I think in 2.2, we said “spec is stable, implementations aren’t there yet”
<emeyer> …if I look at media queries, gird 1 ¶ 2, those are stable specs with hardly any changes and they haven’t made it up to the main thing because the implementations aren’t there
<ChrisL> q?
<emeyer> …the main definition “these are practically Rec and we haven’t done the QA work to verify”
<emeyer> …“Reliable” is for “specs that are implemented with bugs or holes in test suites but very stable specs”
<astearns> ack ChrisL
<fantasai> s/specs that are implemented and very stable (almost no issues)/
<ChrisL> A CSS processor is considered to support a CSS selector if it accepts that all aspects of that selector, recursively, (rather than considering any of its syntax to be unknown or invalid) and that selector doesn’t contain unknown -webkit- pseudo-elements.
<emeyer> ChrisL: This was all nicve theoretical stuff, but I want to come back to the spec
<SebastianZ> q-
<emeyer> …ever since we started the spec, this hasn’t changed and it won’t change
<emeyer> …that’s why 41 tests is reasonable for that sentence
<fantasai> wfm
<emeyer> …so I propose moving this forward to Reliable CR
<emeyer> RESOLVED: css-conditional-for to Reliable CR section