pythongh-99064: Ignore the encoding declaration when parsing decoded text (pythonGH-156734)

ElementTree.parse() with a text file mis-decoded the text in the C
implementation: _parse_whole() encoded it as UTF-8, but left expat to honor
the encoding declared in the document.  It now overrides the encoding, as
XMLParser.feed() already does for str data.
(cherry picked from commit c83013c)

Author: serhiy-storchaka

Lib/test/test_xml_etree.py

@@ -1058,6 +1058,34 @@ def bxml(encoding, body=''):
         self.assertRaises(ValueError, ET.XML, xml('undefined').encode('ascii'))
         self.assertRaises(LookupError, ET.XML, xml('xxx').encode('ascii'))
 
+    def test_parse_text_source(self):
+        # gh-99064: The encoding declared in the document does not apply
+        # to a source which is already decoded.
+        def check(encoding, body):
+            xml = (f"<?xml version='1.0' encoding='{encoding}'?>"
+                   f"<xml>{body}</xml>")
+            with self.subTest(encoding=encoding):
+                self.assertEqual(ET.parse(io.StringIO(xml)).getroot().text,
+                                 body)
+                # the same with an explicitly created parser
+                self.assertEqual(
+                    ET.parse(io.StringIO(xml), ET.XMLParser()).getroot().text,
+                    body)
+        check("ascii", 'a')
+        check("iso-8859-1", '\xbd')
+        check("iso-8859-15", '\u20ac')
+        check("cp437", '\u221a')
+        check("utf-8", '\u4e2d')
+        # not ASCII compatible, unsupported for a bytes source
+        check("utf-16", '\u4e2d')
+        check("utf-32", '\u4e2d')
+
+    def test_parse_text_source_multiple_chunks(self):
+        # the encoding is overridden before the first chunk is parsed
+        body = '\xe4' * 100_000
+        xml = "<?xml version='1.0' encoding='ISO-8859-1'?><xml>%s</xml>" % body
+        self.assertEqual(ET.parse(io.StringIO(xml)).getroot().text, body)
+
     def test_methods(self):
         # Test serialization methods.
 

Misc/NEWS.d/next/Library/2026-08-31-20-30-00.gh-issue-99064.Rt4mZ9.rst

@@ -0,0 +1,5 @@
+Fix :func:`xml.etree.ElementTree.parse` with a text file or other source
+of :class:`str` data in the C implementation.
+The encoding declared in the document was applied to the already decoded
+text, which produced mojibake.  It is now ignored, as when parsing with
+:meth:`!XMLParser.feed` or :func:`~xml.etree.ElementTree.fromstring`.

Modules/_elementtree.c

@@ -4084,6 +4084,7 @@ _elementtree_XMLParser__parse_whole_impl(XMLParserObject *self,
 
     /* read from open file object */
     elementtreestate *st = self->state;
+    int first = 1;
     for (;;) {
 
         buffer = PyObject_CallFunction(reader, "i", 64*1024);
@@ -4100,6 +4101,11 @@ _elementtree_XMLParser__parse_whole_impl(XMLParserObject *self,
                 Py_DECREF(buffer);
                 break;
             }
+            if (first) {
+                /* The text is already decoded, the encoding declared in the
+                   document does not apply to it.  Return code ignored. */
+                (void)EXPAT(st, SetEncoding)(self->parser, "utf-8");
+            }
             temp = PyUnicode_AsEncodedString(buffer, "utf-8", "surrogatepass");
             Py_DECREF(buffer);
             if (!temp) {
@@ -4123,6 +4129,7 @@ _elementtree_XMLParser__parse_whole_impl(XMLParserObject *self,
         res = expat_parse(
             st, self, PyBytes_AS_STRING(buffer), (int)PyBytes_GET_SIZE(buffer),
             0);
+        first = 0;
 
         Py_DECREF(buffer);