diff --git a/Lib/test/test_tools/i18n_data/docstrings.pot b/Lib/test/test_tools/i18n_data/docstrings.pot
index 5af1d41422ff62..387db2413a575f 100644
--- a/Lib/test/test_tools/i18n_data/docstrings.pot
+++ b/Lib/test/test_tools/i18n_data/docstrings.pot
@@ -15,26 +15,40 @@ msgstr ""
 "Generated-By: pygettext.py 1.5\n"
 
 
-#: docstrings.py:7
+#: docstrings.py:1
+#, docstring
+msgid "Module docstring"
+msgstr ""
+
+#: docstrings.py:9
 #, docstring
 msgid ""
 msgstr ""
 
-#: docstrings.py:18
+#: docstrings.py:15
+#, docstring
+msgid "docstring"
+msgstr ""
+
+#: docstrings.py:20
 #, docstring
 msgid ""
 "multiline\n"
-"    docstring\n"
-"    "
+"docstring"
 msgstr ""
 
-#: docstrings.py:25
+#: docstrings.py:27
 #, docstring
 msgid "docstring1"
 msgstr ""
 
-#: docstrings.py:30
+#: docstrings.py:38
+#, docstring
+msgid "nested docstring"
+msgstr ""
+
+#: docstrings.py:43
 #, docstring
-msgid "Hello, {}!"
+msgid "nested class docstring"
 msgstr ""
 
diff --git a/Lib/test/test_tools/i18n_data/docstrings.py b/Lib/test/test_tools/i18n_data/docstrings.py
index 85d7f159d37775..151a55a4b56ba6 100644
--- a/Lib/test/test_tools/i18n_data/docstrings.py
+++ b/Lib/test/test_tools/i18n_data/docstrings.py
@@ -1,3 +1,5 @@
+"""Module docstring"""
+
 # Test docstring extraction
 from gettext import gettext as _
 
@@ -10,10 +12,10 @@ def test(x):
 # Leading empty line
 def test2(x):
 
-    """docstring"""  # XXX This should be extracted but isn't.
+    """docstring"""
 
 
-# XXX Multiline docstrings should be cleaned with `inspect.cleandoc`.
+# Multiline docstrings are cleaned with `inspect.cleandoc`.
 def test3(x):
     """multiline
     docstring
@@ -27,15 +29,15 @@ def test4(x):
 
 
 def test5(x):
-    """Hello, {}!""".format("world!")  # XXX This should not be extracted.
+    """Hello, {}!""".format("world!")  # This should not be extracted.
 
 
 # Nested docstrings
 def test6(x):
     def inner(y):
-        """nested docstring"""  # XXX This should be extracted but isn't.
+        """nested docstring"""
 
 
 class Outer:
     class Inner:
-        "nested class docstring"  # XXX This should be extracted but isn't.
+        "nested class docstring"
diff --git a/Lib/test/test_tools/i18n_data/messages.pot b/Lib/test/test_tools/i18n_data/messages.pot
index 8d66fbc4f3a937..e8167acfc0742b 100644
--- a/Lib/test/test_tools/i18n_data/messages.pot
+++ b/Lib/test/test_tools/i18n_data/messages.pot
@@ -19,22 +19,22 @@ msgstr ""
 msgid ""
 msgstr ""
 
-#: messages.py:19 messages.py:20
+#: messages.py:19 messages.py:20 messages.py:21
 msgid "parentheses"
 msgstr ""
 
-#: messages.py:23
+#: messages.py:24
 msgid "Hello, world!"
 msgstr ""
 
-#: messages.py:26
+#: messages.py:27
 msgid ""
 "Hello,\n"
 "    multiline!\n"
 msgstr ""
 
 #: messages.py:46 messages.py:89 messages.py:90 messages.py:93 messages.py:94
-#: messages.py:99
+#: messages.py:99 messages.py:100 messages.py:101
 msgid "foo"
 msgid_plural "foos"
 msgstr[0] ""
@@ -68,7 +68,7 @@ msgstr ""
 msgid "set"
 msgstr ""
 
-#: messages.py:63
+#: messages.py:62 messages.py:63
 msgid "nested string"
 msgstr ""
 
@@ -76,6 +76,10 @@ msgstr ""
 msgid "baz"
 msgstr ""
 
+#: messages.py:71 messages.py:75
+msgid "default value"
+msgstr ""
+
 #: messages.py:91 messages.py:92 messages.py:95 messages.py:96
 msgctxt "context"
 msgid "foo"
@@ -83,7 +87,13 @@ msgid_plural "foos"
 msgstr[0] ""
 msgstr[1] ""
 
-#: messages.py:100
+#: messages.py:102
 msgid "domain foo"
 msgstr ""
 
+#: messages.py:118 messages.py:119
+msgid "world"
+msgid_plural "worlds"
+msgstr[0] ""
+msgstr[1] ""
+
diff --git a/Lib/test/test_tools/i18n_data/messages.py b/Lib/test/test_tools/i18n_data/messages.py
index 1e03f4e556830d..9457bcb8611020 100644
--- a/Lib/test/test_tools/i18n_data/messages.py
+++ b/Lib/test/test_tools/i18n_data/messages.py
@@ -18,6 +18,7 @@
 # Extra parentheses
 (_("parentheses"))
 ((_("parentheses")))
+_(("parentheses"))
 
 # Multiline strings
 _("Hello, "
@@ -32,7 +33,6 @@
 _(None)
 _(1)
 _(False)
-_(("invalid"))
 _(["invalid"])
 _({"invalid"})
 _("string"[3])
@@ -40,7 +40,7 @@
 _({"string": "foo"})
 
 # pygettext does not allow keyword arguments, but both xgettext and pybabel do
-_(x="kwargs work!")
+_(x="kwargs are not allowed!")
 
 # Unusual, but valid arguments
 _("foo", "bar")
@@ -48,7 +48,7 @@
 
 # .format()
 _("Hello, {}!").format("world")  # valid
-_("Hello, {}!".format("world"))  # invalid, but xgettext and pybabel extract the first string
+_("Hello, {}!".format("world"))  # invalid, but xgettext extracts the first string
 
 # Nested structures
 _("1"), _("2")
@@ -59,7 +59,7 @@
 
 # Nested functions and classes
 def test():
-    _("nested string")  # XXX This should be extracted but isn't.
+    _("nested string")
     [_("nested string")]
 
 
@@ -68,11 +68,11 @@ def bar(self):
         return _("baz")
 
 
-def bar(x=_('default value')):  # XXX This should be extracted but isn't.
+def bar(x=_('default value')):
     pass
 
 
-def baz(x=[_('default value')]):  # XXX This should be extracted but isn't.
+def baz(x=[_('default value')]):
     pass
 
 
@@ -97,6 +97,8 @@ def _(x="don't extract me"):
 
 # Complex arguments
 ngettext("foo", "foos", 42 + (10 - 20))
+ngettext("foo", "foos", *args)
+ngettext("foo", "foos", **kwargs)
 dgettext(["some", {"complex"}, ("argument",)], "domain foo")
 
 # Invalid calls which are not extracted
@@ -108,3 +110,10 @@ def _(x="don't extract me"):
 dngettext('domain', 'foo')
 dpgettext('domain', 'context')
 dnpgettext('domain', 'context', 'foo')
+dgettext(*args, 'foo')
+dpgettext(*args, 'context', 'foo')
+dnpgettext(*args, 'context', 'foo', 'foos')
+
+# f-strings
+f"Hello, {_('world')}!"
+f"Hello, {ngettext('world', 'worlds', 3)}!"
diff --git a/Lib/test/test_tools/test_i18n.py b/Lib/test/test_tools/test_i18n.py
index 29c3423e234d20..d23479104d438d 100644
--- a/Lib/test/test_tools/test_i18n.py
+++ b/Lib/test/test_tools/test_i18n.py
@@ -87,7 +87,7 @@ def assert_POT_equal(self, expected, actual):
         self.maxDiff = None
         self.assertEqual(normalize_POT_file(expected), normalize_POT_file(actual))
 
-    def extract_from_str(self, module_content, *, args=(), strict=True):
+    def extract_from_str(self, module_content, *, args=(), strict=True, with_stderr=False):
         """Return all msgids extracted from module_content."""
         filename = 'test.py'
         with temp_cwd(None):
@@ -98,12 +98,18 @@ def extract_from_str(self, module_content, *, args=(), strict=True):
                 self.assertEqual(res.err, b'')
             with open('messages.pot', encoding='utf-8') as fp:
                 data = fp.read()
-        return self.get_msgids(data)
+        msgids = self.get_msgids(data)
+        if not with_stderr:
+            return msgids
+        return msgids, res.err
 
     def extract_docstrings_from_str(self, module_content):
         """Return all docstrings extracted from module_content."""
         return self.extract_from_str(module_content, args=('--docstrings',), strict=False)
 
+    def get_stderr(self, module_content):
+        return self.extract_from_str(module_content, strict=False, with_stderr=True)[1]
+
     def test_header(self):
         """Make sure the required fields are in the header, according to:
            http://www.gnu.org/software/gettext/manual/gettext.html#Header-Entry
@@ -407,6 +413,24 @@ def test_files_list(self):
             self.assertIn(f'msgid "{text2}"', data)
             self.assertNotIn(text3, data)
 
+    def test_error_messages(self):
+        """Test that pygettext outputs error messages to stderr."""
+        stderr = self.get_stderr(dedent('''\
+        _(1+2)
+        ngettext('foo')
+        dgettext(*args, 'foo')
+        '''))
+
+        # Normalize line endings on Windows
+        stderr = stderr.decode('utf-8').replace('\r', '')
+
+        self.assertEqual(
+            stderr,
+            "*** test.py:1: Expected a string constant for argument 1, got 1 + 2\n"
+            "*** test.py:2: Expected at least 2 positional argument(s) in gettext call, got 1\n"
+            "*** test.py:3: Variable positional arguments are not allowed in gettext calls\n"
+        )
+
 
 def update_POT_snapshots():
     for input_file in DATA_DIR.glob('*.py'):
diff --git a/Misc/NEWS.d/next/Tools-Demos/2023-05-11-23-32-25.gh-issue-104400.23vxm7.rst b/Misc/NEWS.d/next/Tools-Demos/2023-05-11-23-32-25.gh-issue-104400.23vxm7.rst
new file mode 100644
index 00000000000000..82954e170ec3e2
--- /dev/null
+++ b/Misc/NEWS.d/next/Tools-Demos/2023-05-11-23-32-25.gh-issue-104400.23vxm7.rst
@@ -0,0 +1 @@
+Fix several bugs in extraction by switching to an AST parser in :program:`pygettext`.
diff --git a/Tools/i18n/pygettext.py b/Tools/i18n/pygettext.py
index d8a0e379ab82cb..4720aecbdc515a 100755
--- a/Tools/i18n/pygettext.py
+++ b/Tools/i18n/pygettext.py
@@ -140,8 +140,6 @@
 import os
 import sys
 import time
-import tokenize
-from collections import defaultdict
 from dataclasses import dataclass, field
 from operator import itemgetter
 
@@ -206,15 +204,6 @@ def escape_nonascii(s, encoding):
     return ''.join(escapes[b] for b in s.encode(encoding))
 
 
-def is_literal_string(s):
-    return s[0] in '\'"' or (s[0] in 'rRuU' and s[1] in '\'"')
-
-
-def safe_eval(s):
-    # unwrap quotes, safely
-    return eval(s, {'__builtins__':{}}, {})
-
-
 def normalize(s, encoding):
     # This converts the various Python string types into a format that is
     # appropriate for .po files, namely much closer to C style.
@@ -296,11 +285,6 @@ def getFilesForName(name):
 }
 
 
-def matches_spec(message, spec):
-    """Check if a message has all the keys defined by the keyword spec."""
-    return all(key in message for key in spec.values())
-
-
 @dataclass(frozen=True)
 class Location:
     filename: str
@@ -325,203 +309,91 @@ def add_location(self, filename, lineno, msgid_plural=None, *, is_docstring=Fals
         self.is_docstring |= is_docstring
 
 
-class TokenEater:
+class GettextVisitor(ast.NodeVisitor):
     def __init__(self, options):
-        self.__options = options
-        self.__messages = {}
-        self.__state = self.__waiting
-        self.__data = defaultdict(str)
-        self.__curr_arg = 0
-        self.__curr_keyword = None
-        self.__lineno = -1
-        self.__freshmodule = 1
-        self.__curfile = None
-        self.__enclosurecount = 0
-
-    def __call__(self, ttype, tstring, stup, etup, line):
-        # dispatch
-##        import token
-##        print('ttype:', token.tok_name[ttype], 'tstring:', tstring,
-##              file=sys.stderr)
-        self.__state(ttype, tstring, stup[0])
-
-    @property
-    def messages(self):
-        return self.__messages
-
-    def __waiting(self, ttype, tstring, lineno):
-        opts = self.__options
-        # Do docstring extractions, if enabled
-        if opts.docstrings and not opts.nodocstrings.get(self.__curfile):
-            # module docstring?
-            if self.__freshmodule:
-                if ttype == tokenize.STRING and is_literal_string(tstring):
-                    self.__addentry({'msgid': safe_eval(tstring)}, lineno, is_docstring=True)
-                    self.__freshmodule = 0
-                    return
-                if ttype in (tokenize.COMMENT, tokenize.NL, tokenize.ENCODING):
-                    return
-                self.__freshmodule = 0
-            # class or func/method docstring?
-            if ttype == tokenize.NAME and tstring in ('class', 'def'):
-                self.__state = self.__suiteseen
-                return
-        if ttype == tokenize.NAME and tstring in ('class', 'def'):
-            self.__state = self.__ignorenext
+        super().__init__()
+        self.options = options
+        self.filename = None
+        self.messages = {}
+
+    def visit_file(self, node, filename):
+        self.filename = filename
+        self.visit(node)
+
+    def visit_Module(self, node):
+        self._extract_docstring(node)
+        self.generic_visit(node)
+
+    visit_ClassDef = visit_FunctionDef = visit_AsyncFunctionDef = visit_Module
+
+    def visit_Call(self, node):
+        self._extract_message(node)
+        self.generic_visit(node)
+
+    def _extract_docstring(self, node):
+        if (not self.options.docstrings or
+            self.options.nodocstrings.get(self.filename)):
             return
-        if ttype == tokenize.NAME and tstring in opts.keywords:
-            self.__state = self.__keywordseen
-            self.__curr_keyword = tstring
+
+        docstring = ast.get_docstring(node)
+        if docstring is not None:
+            lineno = node.body[0].lineno  # The first statement is the docstring
+            self._add_message(lineno, docstring, is_docstring=True)
+
+    def _extract_message(self, node):
+        func_name = self._get_func_name(node)
+        spec = self.options.keywords.get(func_name)
+        if spec is None:
+            return
+
+        max_index = max(spec)
+        has_var_positional = any(isinstance(arg, ast.Starred) for
+                                 arg in node.args[:max_index+1])
+        if has_var_positional:
+            print(f'*** {self.filename}:{node.lineno}: Variable positional '
+                  f'arguments are not allowed in gettext calls', file=sys.stderr)
             return
-        if ttype == tokenize.STRING:
-            maybe_fstring = ast.parse(tstring, mode='eval').body
-            if not isinstance(maybe_fstring, ast.JoinedStr):
-                return
-            for value in filter(lambda node: isinstance(node, ast.FormattedValue),
-                                maybe_fstring.values):
-                for call in filter(lambda node: isinstance(node, ast.Call),
-                                   ast.walk(value)):
-                    func = call.func
-                    if isinstance(func, ast.Name):
-                        func_name = func.id
-                    elif isinstance(func, ast.Attribute):
-                        func_name = func.attr
-                    else:
-                        continue
-
-                    if func_name not in opts.keywords:
-                        continue
-                    if len(call.args) != 1:
-                        print((
-                            '*** %(file)s:%(lineno)s: Seen unexpected amount of'
-                            ' positional arguments in gettext call: %(source_segment)s'
-                            ) % {
-                            'source_segment': ast.get_source_segment(tstring, call) or tstring,
-                            'file': self.__curfile,
-                            'lineno': lineno
-                            }, file=sys.stderr)
-                        continue
-                    if call.keywords:
-                        print((
-                            '*** %(file)s:%(lineno)s: Seen unexpected keyword arguments'
-                            ' in gettext call: %(source_segment)s'
-                            ) % {
-                            'source_segment': ast.get_source_segment(tstring, call) or tstring,
-                            'file': self.__curfile,
-                            'lineno': lineno
-                            }, file=sys.stderr)
-                        continue
-                    arg = call.args[0]
-                    if not isinstance(arg, ast.Constant):
-                        print((
-                            '*** %(file)s:%(lineno)s: Seen unexpected argument type'
-                            ' in gettext call: %(source_segment)s'
-                            ) % {
-                            'source_segment': ast.get_source_segment(tstring, call) or tstring,
-                            'file': self.__curfile,
-                            'lineno': lineno
-                            }, file=sys.stderr)
-                        continue
-                    if isinstance(arg.value, str):
-                        self.__curr_keyword = func_name
-                        self.__addentry({'msgid': arg.value}, lineno)
-
-    def __suiteseen(self, ttype, tstring, lineno):
-        # skip over any enclosure pairs until we see the colon
-        if ttype == tokenize.OP:
-            if tstring == ':' and self.__enclosurecount == 0:
-                # we see a colon and we're not in an enclosure: end of def
-                self.__state = self.__suitedocstring
-            elif tstring in '([{':
-                self.__enclosurecount += 1
-            elif tstring in ')]}':
-                self.__enclosurecount -= 1
-
-    def __suitedocstring(self, ttype, tstring, lineno):
-        # ignore any intervening noise
-        if ttype == tokenize.STRING and is_literal_string(tstring):
-            self.__addentry({'msgid': safe_eval(tstring)}, lineno, is_docstring=True)
-            self.__state = self.__waiting
-        elif ttype not in (tokenize.NEWLINE, tokenize.INDENT,
-                           tokenize.COMMENT):
-            # there was no class docstring
-            self.__state = self.__waiting
-
-    def __keywordseen(self, ttype, tstring, lineno):
-        if ttype == tokenize.OP and tstring == '(':
-            self.__data.clear()
-            self.__curr_arg = 0
-            self.__enclosurecount = 0
-            self.__lineno = lineno
-            self.__state = self.__openseen
-        else:
-            self.__state = self.__waiting
-
-    def __openseen(self, ttype, tstring, lineno):
-        spec = self.__options.keywords[self.__curr_keyword]
-        arg_type = spec.get(self.__curr_arg)
-        expect_string_literal = arg_type is not None
-
-        if ttype == tokenize.OP and self.__enclosurecount == 0:
-            if tstring == ')':
-                # We've seen the last of the translatable strings.  Record the
-                # line number of the first line of the strings and update the list
-                # of messages seen.  Reset state for the next batch.  If there
-                # were no strings inside _(), then just ignore this entry.
-                if self.__data:
-                    self.__addentry(self.__data)
-                self.__state = self.__waiting
-                return
-            elif tstring == ',':
-                # Advance to the next argument
-                self.__curr_arg += 1
-                return
 
-        if expect_string_literal:
-            if ttype == tokenize.STRING and is_literal_string(tstring):
-                self.__data[arg_type] += safe_eval(tstring)
-            elif ttype not in (tokenize.COMMENT, tokenize.INDENT, tokenize.DEDENT,
-                               tokenize.NEWLINE, tokenize.NL):
-                # We are inside an argument which is a translatable string and
-                # we encountered a token that is not a string.  This is an error.
-                self.warn_unexpected_token(tstring)
-                self.__enclosurecount = 0
-                self.__state = self.__waiting
-        elif ttype == tokenize.OP:
-            if tstring in '([{':
-                self.__enclosurecount += 1
-            elif tstring in ')]}':
-                self.__enclosurecount -= 1
-
-    def __ignorenext(self, ttype, tstring, lineno):
-        self.__state = self.__waiting
-
-    def __addentry(self, msg, lineno=None, *, is_docstring=False):
-        msgid = msg.get('msgid')
-        if msgid in self.__options.toexclude:
+        if max_index >= len(node.args):
+            print(f'*** {self.filename}:{node.lineno}: Expected at least '
+                  f'{max(spec) + 1} positional argument(s) in gettext call, '
+                  f'got {len(node.args)}', file=sys.stderr)
             return
-        if not is_docstring:
-            spec = self.__options.keywords[self.__curr_keyword]
-            if not matches_spec(msg, spec):
+
+        msg_data = {}
+        for position, arg_type in spec.items():
+            arg = node.args[position]
+            if not self._is_string_const(arg):
+                print(f'*** {self.filename}:{arg.lineno}: Expected a string '
+                      f'constant for argument {position + 1}, '
+                      f'got {ast.unparse(arg)}', file=sys.stderr)
                 return
-        if lineno is None:
-            lineno = self.__lineno
-        msgctxt = msg.get('msgctxt')
-        msgid_plural = msg.get('msgid_plural')
+            msg_data[arg_type] = arg.value
+
+        lineno = node.lineno
+        self._add_message(lineno, **msg_data)
+
+    def _add_message(
+            self, lineno, msgid, msgid_plural=None, msgctxt=None, *,
+            is_docstring=False):
+        if msgid in self.options.toexclude:
+            return
+
         key = self._key_for(msgid, msgctxt)
-        if key in self.__messages:
-            self.__messages[key].add_location(
-                self.__curfile,
+        message = self.messages.get(key)
+        if message:
+            message.add_location(
+                self.filename,
                 lineno,
                 msgid_plural,
                 is_docstring=is_docstring,
             )
         else:
-            self.__messages[key] = Message(
+            self.messages[key] = Message(
                 msgid=msgid,
                 msgid_plural=msgid_plural,
                 msgctxt=msgctxt,
-                locations={Location(self.__curfile, lineno)},
+                locations={Location(self.filename, lineno)},
                 is_docstring=is_docstring,
             )
 
@@ -531,19 +403,17 @@ def _key_for(msgid, msgctxt=None):
             return (msgctxt, msgid)
         return msgid
 
-    def warn_unexpected_token(self, token):
-        print((
-            '*** %(file)s:%(lineno)s: Seen unexpected token "%(token)s"'
-            ) % {
-            'token': token,
-            'file': self.__curfile,
-            'lineno': self.__lineno
-            }, file=sys.stderr)
-
-    def set_filename(self, filename):
-        self.__curfile = filename
-        self.__freshmodule = 1
+    def _get_func_name(self, node):
+        match node.func:
+            case ast.Name(id=id):
+                return id
+            case ast.Attribute(attr=attr):
+                return attr
+            case _:
+                return None
 
+    def _is_string_const(self, node):
+        return isinstance(node, ast.Constant) and isinstance(node.value, str)
 
 def write_pot_file(messages, options, fp):
     timestamp = time.strftime('%Y-%m-%d %H:%M%z')
@@ -717,31 +587,24 @@ class Options:
     args = expanded
 
     # slurp through all the files
-    eater = TokenEater(options)
+    visitor = GettextVisitor(options)
     for filename in args:
         if filename == '-':
             if options.verbose:
                 print('Reading standard input')
-            fp = sys.stdin.buffer
-            closep = 0
+            source = sys.stdin.buffer.read()
         else:
             if options.verbose:
                 print(f'Working on {filename}')
-            fp = open(filename, 'rb')
-            closep = 1
+            with open(filename, 'rb') as fp:
+                source = fp.read()
+
         try:
-            eater.set_filename(filename)
-            try:
-                tokens = tokenize.tokenize(fp.readline)
-                for _token in tokens:
-                    eater(*_token)
-            except tokenize.TokenError as e:
-                print('%s: %s, line %d, column %d' % (
-                    e.args[0], filename, e.args[1][0], e.args[1][1]),
-                    file=sys.stderr)
-        finally:
-            if closep:
-                fp.close()
+            module_tree = ast.parse(source)
+        except SyntaxError:
+            continue
+
+        visitor.visit_file(module_tree, filename)
 
     # write the output
     if options.outfile == '-':
@@ -753,7 +616,7 @@ class Options:
         fp = open(options.outfile, 'w')
         closep = 1
     try:
-        write_pot_file(eater.messages, options, fp)
+        write_pot_file(visitor.messages, options, fp)
     finally:
         if closep:
             fp.close()