Fix GH-20281: \Dom\Document::getElementById() is inconsistent after nodes are removed

This worked for non-parsed elements already, but not for elements where
xmlAddID() returns early due to the ID already existing.
In that case what was missing is marking the attribute as an ID.

Closes GH-20283.

Author: ndossche

NEWS

@@ -15,6 +15,8 @@ PHP                                                                        NEWS
 - DOM:
   . Partially fixed bug GH-16317 (DOM classes do not allow
     __debugInfo() overrides to work). (nielsdos)
+  . Fixed bug GH-20281 (\Dom\Document::getElementById() is inconsistent
+    after nodes are removed). (nielsdos)
 
 - Exif:
   . Fix possible memory leak when tag is empty. (nielsdos)

ext/dom/html5_parser.c

@@ -268,7 +268,10 @@ static lexbor_libxml2_bridge_status lexbor_libxml2_bridge_convert(
 
                 /* xmlIsID does some other stuff too that is irrelevant here. */
                 if (local_name_length == 2 && local_name[0] == 'i' && local_name[1] == 'd' && attr->node.ns == LXB_NS_HTML) {
-                    xmlAddID(NULL, lxml_doc, value, lxml_attr);
+                    if (xmlAddID(NULL, lxml_doc, value, lxml_attr) == 0) {
+                        /* If the ID already exists, the ID attribute still needs to be marked as an ID. */
+                        lxml_attr->atype = XML_ATTRIBUTE_ID;
+                    }
                 }
 
                 /* libxml2 doesn't support line numbers on this anyway, it derives them instead, so don't bother */

ext/dom/tests/modern/html/interactions/gh20281.phpt

@@ -0,0 +1,15 @@
+--TEST--
+GH-20281 (\Dom\Document::getElementById() is inconsistent after nodes are removed)
+--EXTENSIONS--
+dom
+--CREDITS--
+cscott
+--FILE--
+<?php
+$d = \Dom\HTMLDocument::createFromString('<p id="a">b</p><p id="a">c</p>', LIBXML_NOERROR);
+$p = $d->getElementById('a');
+$p->remove();
+echo $d->getElementById('a')->textContent, "\n";
+?>
+--EXPECT--
+c