Selector: Stop relying on CSS.supports( "selector(...)" )

`CSS.supports( "selector(...)" )` has different semantics than selectors passed
to `querySelectorAll`. Apart from the fact that the former returns `false` for
unrecognized selectors and the latter throws, `qSA` is more forgiving and
accepts some invalid selectors, auto-correcting them where needed - for
example, mismatched brackers are auto-closed. This behavior difference is
breaking for many users.

To add to that, a recent CSSWG resolution made `:is()` & `:where()` the only
pseudos with forgiving parsing; browsers are in the process of making `:has()`
parsing unforgiving.

Taking all that into account, we go back to our previous try-catch approach
without relying on `CSS.supports( "selector(...)" )`. The only difference
is we detect forgiving parsing in `:has()` and mark the selector as buggy.

Fixes jquery/jquery#5194
Closes gh-493
Ref jquery/jquery#5098
Ref jquery/jquery#5206
Ref jquery/jquery#5207
Ref gh-486
Ref w3c/csswg-drafts#7676

src/sizzle.js

@@ -356,27 +356,6 @@ function Sizzle( selector, context, results, seed ) {
 				}
 
 				try {
-
-					// `qSA` may not throw for unrecognized parts using forgiving parsing:
-					// https://drafts.csswg.org/selectors/#forgiving-selector
-					// like the `:has()` pseudo-class:
-					// https://drafts.csswg.org/selectors/#relational
-					// `CSS.supports` is still expected to return `false` then:
-					// https://drafts.csswg.org/css-conditional-4/#typedef-supports-selector-fn
-					// https://drafts.csswg.org/css-conditional-4/#dfn-support-selector
-					if ( support.cssSupportsSelector &&
-
-						// eslint-disable-next-line no-undef
-						!CSS.supports( "selector(:is(" + newSelector + "))" ) ) {
-
-						// Support: IE 11+
-						// Throw to get to the same code path as an error directly in qSA.
-						// Note: once we only support browser supporting
-						// `CSS.supports('selector(...)')`, we can most likely drop
-						// the `try-catch`. IE doesn't implement the API.
-						throw new Error();
-					}
-
 					push.apply( results,
 						newContext.querySelectorAll( newSelector )
 					);
@@ -672,29 +651,22 @@ setDocument = Sizzle.setDocument = function( node ) {
 			!el.querySelectorAll( ":scope fieldset div" ).length;
 	} );
 
-	// Support: Chrome 105+, Firefox 104+, Safari 15.4+
-	// Make sure forgiving mode is not used in `CSS.supports( "selector(...)" )`.
-	//
-	// `:is()` uses a forgiving selector list as an argument and is widely
-	// implemented, so it's a good one to test against.
-	support.cssSupportsSelector = assert( function() {
-		/* eslint-disable no-undef */
-
-		return CSS.supports( "selector(*)" ) &&
-
-			// Support: Firefox 78-81 only
-			// In old Firefox, `:is()` didn't use forgiving parsing. In that case,
-			// fail this test as there's no selector to test against that.
-			// `CSS.supports` uses unforgiving parsing
-			document.querySelectorAll( ":is(:jqfake)" ) &&
-
-			// `*` is needed as Safari & newer Chrome implemented something in between
-			// for `:has()` - it throws in `qSA` if it only contains an unsupported
-			// argument but multiple ones, one of which is supported, are fine.
-			// We want to play safe in case `:is()` gets the same treatment.
-			!CSS.supports( "selector(:is(*,:jqfake))" );
-
-		/* eslint-enable */
+	// Support: Chrome 105 - 110+, Safari 15.4 - 16.3+
+	// Make sure the the `:has()` argument is parsed unforgivingly.
+	// We include `*` in the test to detect buggy implementations that are
+	// _selectively_ forgiving (specifically when the list includes at least
+	// one valid selector).
+	// Note that we treat complete lack of support for `:has()` as if it were
+	// spec-compliant support, which is fine because use of `:has()` in such
+	// environments will fail in the qSA path and fall back to jQuery traversal
+	// anyway.
+	support.cssHas = assert( function() {
+		try {
+			document.querySelector( ":has(*,:jqfake)" );
+			return false;
+		} catch ( e ) {
+			return true;
+		}
 	} );
 
 	/* Attributes
@@ -963,14 +935,14 @@ setDocument = Sizzle.setDocument = function( node ) {
 		} );
 	}
 
-	if ( !support.cssSupportsSelector ) {
+	if ( !support.cssHas ) {
 
-		// Support: Chrome 105+, Safari 15.4+
-		// `:has()` uses a forgiving selector list as an argument so our regular
-		// `try-catch` mechanism fails to catch `:has()` with arguments not supported
-		// natively like `:has(:contains("Foo"))`. Where supported & spec-compliant,
-		// we now use `CSS.supports("selector(:is(SELECTOR_TO_BE_TESTED))")`, but
-		// outside that we mark `:has` as buggy.
+		// Support: Chrome 105 - 110+, Safari 15.4 - 16.3+
+		// Our regular `try-catch` mechanism fails to detect natively-unsupported
+		// pseudo-classes inside `:has()` (such as `:has(:contains("Foo"))`)
+		// in browsers that parse the `:has()` argument as a forgiving selector list.
+		// https://drafts.csswg.org/selectors/#relational now requires the argument
+		// to be parsed unforgivingly, but browsers have not yet fully adjusted.
 		rbuggyQSA.push( ":has" );
 	}