fix(compiler): answer an ancestor condition that names no class

[YevheniiKotyrlo]

Problem

A descendant combinator opens a container-query ref, and every later component of that compound writes its condition onto it. Only the class arm attaches the ref to the rule, so an ancestor identified by a condition alone is discarded and the rule applies to every element the class names, in every state. The container queries each rule compiles to on main:

selector	main	here
[data-state="on"] .x	none, so it applies in every state	[{"a":[["d","state","=","on"]]}]
:hover .x	none	[{"p":{"h":1}}]
:disabled .x	none	[{"a":[["a","disabled"]]}]
.g[data-state="on"] .x	[{"a":[…],"n":"g:g"}]	unchanged
.x:where([data-state="on"] *)	[{"a":[["d","state","=","on"]]}]	unchanged

The last two rows are why this is a defect rather than a limit. The class arm attaches a named query and parseIsWhereComponents already emits an unnamed one for the :where(… *) spelling, so two spellings of one selector disagree, and the one an author writes by hand fails open. A descendant combinator asks for an ancestor matching the whole compound to its left (Selectors 4 §14.1), so a condition in that compound is part of the match.

Solution

The ref is attached when a condition is written, not only when a class names it. attachContainerQuery adds a ref once, so a compound carrying both a class and a condition attaches one query and the named path compiles as before.

The condition then resolves against the nearest container, as the :where() spelling does. With no container above, testContainerQuery answers false, so a selector the runtime cannot answer fails closed.

Tests

• src/__tests__/compiler/ancestor-conditions.test.ts: each row above, a chain of two classless ancestors, a named and a classless ancestor together, a multi-class compound yielding one query, the combinator compiling to what :where() does, a compound on one element staying the subject's, and an unanswerable combinator still dropping the rule.
• src/__tests__/native/ancestor-conditions.test.tsx, through the runtime:
  • hover and press withheld until the container is hovered or pressed
  • [data-state="on"] and :disabled ancestors following the container's props both ways
  • withheld where no ancestor can answer and from a sibling's subtree
  • a named ancestor still answering from its own element

On main, 14 of the 22 cases in the two files fail; all pass here. Removing the attach from the hover, attribute and disabled arms fails their own cases.

Every ancestor-state test in the suite names a class, because that is what Tailwind emits: group-hover: compiles to .group:hover .x, which was already correct. The classless form is what an author writes in a plain stylesheet.

Verification

On Windows with Node 26:

• yarn lint clean
• yarn typecheck clean
• yarn test --maxWorkers=2 --coverage: 1353 passed, 4 failed. The 4 are the babel cases that also fail on main on this machine.
• yarn build clean
• yarn example expo export --platform web exported
• nothing unstaged after either build

No existing issue tracks this. I searched for ancestor attribute selector, data- descendant, hover descendant, group-hover and parent hover.

Device evidence, before and after: #462 (comment)

Known limits

An unnamed query resolves the nearest ancestor that established a container, not any ancestor. That under-matches, as :where(… *) already does. Going further needs every element to be addressable, and the stylesheet is class-keyed: an element carrying no class the sheet names registers nothing.

Independent of #463, which decides the order of an ancestor chain at runtime. They touch no common file.

Merge order

It shares lines with #448 (compiler/selector-builder.ts), #459 (compiler/selector-builder.ts); whichever lands second rebases.

Base

Re-written on main (a5002c5).

[YevheniiKotyrlo]

Device evidence — before / after

UNFIXED — All three bars are red. The classless ancestor’s condition is written onto a ref nothing reads, so the rule compiles with no condition and applies under every ancestor.

FIXED — Only the first bar is red. [data-state="on"] .probe applies under the ancestor that carries the attribute, and nowhere else.

before — 3.1.0-rc.0	after — with this PR

Android 16 emulator (API 36, GLES on the host GPU), 1140×2400 @ 480dpi, both frames from one run. before is the published 3.1.0-rc.0 plus what my test app needs from my other patches to run: the native and native-internal subpath exports, PR 459, cascade-layer ranking, and a shim that compiles a class-qualified :root inert. after adds the rest of my open PRs, this one included. One of them shows on every screen: PR 417 narrows a var()-valued font-family to one family, so the after sets its text in the theme's face where the before falls back to the platform's.