Follow spec so < can start an attribute name

Fixes #1483

CHANGES.md

@@ -39,6 +39,9 @@
   created (`html` or `body`). [2204](https://github.com/jhy/jsoup/issues/2204)
 * Follow the current HTML specification in the tokenizer to allow `<` as part of a tag name, instead of emitting it as a
   character node. [2230](https://github.com/jhy/jsoup/issues/2230)
+* Similarly, allow a `<` as the start of an attribute name, vs creating a new element. The previous behavior was
+  intended to parse closer to what we anticipated the author's intent to be, but that does not align to the spec or to
+  how browsers behave. [1483](https://github.com/jhy/jsoup/issues/1483)
 
 ## 1.18.1 (2024-Jul-10)
 

src/main/java/org/jsoup/parser/TokeniserState.java

@@ -568,10 +568,6 @@ private void anythingElse(Tokeniser t, CharacterReader r) {
                 case '/':
                     t.transition(SelfClosingStartTag);
                     break;
-                case '<': // NOTE: out of spec, but clear (spec has this as a part of the attribute name)
-                    r.unconsume();
-                    t.error(this);
-                    // intended fall through as if >
                 case '>':
                     t.emitTagPending();
                     t.transition(Data);

src/test/java/org/jsoup/parser/HtmlParserTest.java

@@ -79,13 +79,13 @@ private static Stream<Arguments> dupeAttributeData() {
 
     @Test public void parsesQuiteRoughAttributes() {
         String html = "<p =a>One<a <p>Something</p>Else";
-        // this (used to; now gets cleaner) gets a <p> with attr '=a' and an <a tag with an attribute named '<p'; and then auto-recreated
+        // this gets a <p> with attr '=a' and an <a tag with an attribute named '<p'; and then auto-recreated
         Document doc = Jsoup.parse(html);
 
-        // NOTE: per spec this should be the test case. but impacts too many ppl
-        // assertEquals("<p =a>One<a <p>Something</a></p>\n<a <p>Else</a>", doc.body().html());
-
-        assertEquals("<p _a>One<a></a></p><p><a>Something</a></p><a>Else</a>", TextUtil.stripNewlines(doc.body().html()));
+        // =a is output as _a
+        assertEquals("<p _a>One<a <p>Something</a></p><a <p>Else</a>", TextUtil.stripNewlines(doc.body().html()));
+        Element p = doc.expectFirst("p");
+        assertNotNull(p.attribute("=a"));
 
         doc = Jsoup.parse("<p .....>");
         assertEquals("<p .....></p>", doc.body().html());
@@ -1939,4 +1939,17 @@ private static void assertMathNamespace(Element el) {
         assertEquals("Hello", ab.text());
         assertEquals("a<b", ab.tag().normalName());
     }
+
+    @Test void ltInAttrStart() {
+        // https://github.com/jhy/jsoup/issues/1483
+        String html = "<a before='foo' <junk after='bar'>One</a>";
+        Document doc = Jsoup.parse(html);
+        assertEquals("<a before=\"foo\" <junk after=\"bar\">One</a>", TextUtil.normalizeSpaces(doc.body().html()));
+
+        Element el = doc.expectFirst("a");
+        Attribute attribute = el.attribute("<junk");
+        assertNotNull(attribute);
+        assertEquals("", attribute.getValue());
+
+    }
 }

src/test/java/org/jsoup/parser/TokeniserStateTest.java

@@ -230,7 +230,7 @@ public void testOpeningAngleBracketInsteadOfAttribute() {
 
         Parser.parseFragment(triggeringSnippet, null, "", errorList);
 
-        assertEquals(6, errorList.get(0).getPosition());
+        assertEquals(7, errorList.get(0).getPosition());
     }
 
     @Test