gh-149277: Fix error position for invalid numeric literals

Author: skirpichev

Lib/test/test_grammar.py

@@ -21,6 +21,7 @@
     import_helper,
     skip_emscripten_stack_overflow,
     skip_wasi_stack_overflow,
+    subTests,
 )
 from test.support.numbers import (
     VALID_UNDERSCORE_LITERALS,
@@ -180,6 +181,27 @@ def check(test, error=False):
         check("[0x1for x in ()]")
         check("[0xfor x in ()]")
 
+    @subTests('source,offset,msg',
+              [("0xfg", 4, "hexadecimal"),
+               ("0x9g", 4, "hexadecimal"),
+               ("0b1z", 4, "binary"),
+               ("0o7q", 4, "octal"),
+               ("9spam", 2, " decimal"),
+               ("0xfspam", 4, "hexadecimal"),
+               ("1.0x", 4, " decimal"),
+               ("1e3w", 4, " decimal"),
+               ("1jz", 3, "imaginary"),
+               ("0xI", 3, "hexadecimal"),
+               ("0bz", 3, "binary"),
+               ])
+    def test_end_of_numerical_literals_offset(self, source, offset, msg):
+        # gh-149277: verify the error caret points at the first invalid
+        # character, not the last valid digit.
+        with self.assertRaises(SyntaxError) as cm:
+            compile(source, "<test>", "eval")
+        self.assertEqual(cm.exception.offset, offset)
+        self.assertIn(msg, cm.exception.msg)
+
     def test_string_literals(self):
         x = ''; y = ""; self.assertTrue(len(x) == 0 and x == y)
         x = '\''; y = "'"; self.assertTrue(len(x) == 1 and x == y and ord(x) == 39)

Lib/test/test_tokenize.py

@@ -2277,7 +2277,7 @@ def test_extra_tokens_relaxes_lexer_errors(self):
         cases = [
             (
                 "2sin(x)",
-                ("invalid decimal literal", (1, 1)),
+                ("invalid decimal literal", (1, 2)),
                 [
                     (token.NUMBER, "2", (1, 0), (1, 1)),
                     (token.NAME, "sin", (1, 1), (1, 4)),

Misc/NEWS.d/next/Core_and_Builtins/2026-08-11-08-16-20.gh-issue-149277.dYRtaF.rst

@@ -0,0 +1,2 @@
+Fix the :exc:`SyntaxError` caret position for invalid numeric literals to
+point at the first invalid character.

Parser/lexer/number.c

@@ -80,7 +80,6 @@ verify_end_of_number(struct tok_state *tok, int c, const char *kind) {
     }
     else /* In future releases, only error will remain. */
     if (c < 128 && is_potential_identifier_char(c)) {
-        tok_backup(tok, c);
         _PyTokenizer_syntaxerror(tok, "invalid %s literal", kind);
         return 0;
     }
@@ -130,7 +129,6 @@ _PyLexer_scan_number(struct tok_state *tok, struct token *token, int c,
                     c = tok_nextc(tok);
                 }
                 if (!Py_ISXDIGIT(c)) {
-                    tok_backup(tok, c);
                     return MAKE_TOKEN(_PyTokenizer_syntaxerror(tok, "invalid hexadecimal literal"));
                 }
                 do {
@@ -154,7 +152,6 @@ _PyLexer_scan_number(struct tok_state *tok, struct token *token, int c,
                                 "invalid digit '%c' in octal literal", c));
                     }
                     else {
-                        tok_backup(tok, c);
                         return MAKE_TOKEN(_PyTokenizer_syntaxerror(tok, "invalid octal literal"));
                     }
                 }
@@ -182,7 +179,6 @@ _PyLexer_scan_number(struct tok_state *tok, struct token *token, int c,
                         return MAKE_TOKEN(_PyTokenizer_syntaxerror(tok, "invalid digit '%c' in binary literal", c));
                     }
                     else {
-                        tok_backup(tok, c);
                         return MAKE_TOKEN(_PyTokenizer_syntaxerror(tok, "invalid binary literal"));
                     }
                 }