Fix normalize-space() collapsing NBSP (U+00A0) into a regular space

src/expressions/function-call-expression.ts

@@ -69,7 +69,10 @@ const BUILT_IN_FUNCTIONS: Record<string, (context: XPathContext, ...args: any[])
         return s.substring(adjustedStart, adjustedStart + length);
     },
     'string-length': (_ctx, arg) => String(arg).length,
-    'normalize-space': (_ctx, arg) => String(arg).trim().replace(/\s+/g, ' '),
+    'normalize-space': (_ctx, arg) =>
+        String(arg)
+            .replace(/^[\t\n\r ]+|[\t\n\r ]+$/g, '')
+            .replace(/[\t\n\r ]+/g, ' '),
     contains: (_ctx, str, sub) => String(str).includes(String(sub)),
     'starts-with': (_ctx, str, sub) => String(str).startsWith(String(sub)),
     'ends-with': (_ctx, str, sub) => String(str).endsWith(String(sub)),
@@ -935,7 +938,11 @@ export class XPathFunctionCall extends XPathExpression {
 
     private normalizeSpace(args: XPathResult[], context: XPathContext): string {
         const str = args.length === 0 ? this.stringValue([], context) : this.convertToString(args[0]);
-        return str.trim().replace(/\s+/g, ' ');
+        // Per the XPath/XQuery F&O spec, normalize-space() only collapses XML whitespace
+        // (#x9, #xA, #xD, #x20). Using JS's .trim()/\s here is wrong because ECMA-262
+        // treats U+00A0 (NBSP) and other Unicode spaces as whitespace too, which silently
+        // corrupts real content (e.g. NBSP in text) into a plain space.
+        return str.replace(/^[\t\n\r ]+|[\t\n\r ]+$/g, '').replace(/[\t\n\r ]+/g, ' ');
     }
 
     private substringBefore(args: XPathResult[]): string {

tests/expressions/expressions.test.ts

@@ -549,6 +549,26 @@ describe('Expression Evaluation', () => {
             expect(result).toBe('text');
         });
 
+        it('should preserve non-breaking spaces (U+00A0) instead of collapsing them to a regular space', () => {
+            // Per the XPath/XQuery F&O spec, normalize-space() only treats #x9/#xA/#xD/#x20 as
+            // whitespace. U+00A0 (NBSP) is a distinct character and must be left untouched -
+            // it must not be collapsed or trimmed away like ordinary XML whitespace.
+            const result = evaluate("normalize-space('Foo Bar')");
+            expect(result).toBe('Foo Bar');
+        });
+
+        it('should trim surrounding XML whitespace while preserving an internal NBSP run', () => {
+            const result = evaluate("normalize-space('  Foo  Bar  ')");
+            expect(result).toBe('Foo  Bar');
+        });
+
+        it('should evaluate normalize-space without argument preserving NBSP in node text', () => {
+            const result = evaluate('normalize-space()', {
+                node: { textContent: '  Foo Bar  ', nodeType: 1, nodeName: 'test' } as any,
+            });
+            expect(result).toBe('Foo Bar');
+        });
+
         it('should evaluate translate', () => {
             const result = evaluate("translate('hello', 'el', 'ip')");
             expect(result).toBe('hippo');