Issue #20387: Merge

Lib/test/test_tokenize.py

@@ -5,6 +5,8 @@ The tests can be really simple. Given a small fragment of source
 code, print out a table with tokens. The ENDMARKER is omitted for
 brevity.
 
+    >>> import glob
+
     >>> dump_tokens("1 + 1")
     ENCODING   'utf-8'       (0, 0) (0, 0)
     NUMBER     '1'           (1, 0) (1, 1)
@@ -647,7 +649,7 @@ from tokenize import (tokenize, _tokenize, untokenize, NUMBER, NAME, OP,
                      open as tokenize_open, Untokenizer)
 from io import BytesIO
 from unittest import TestCase, mock
-import os, sys, glob
+import os
 import token
 
 def dump_tokens(s):
@@ -1227,6 +1229,22 @@ class UntokenizeTest(TestCase):
         self.assertEqual(untokenize(iter(tokens)), b'Hello ')
 
 
+class TestRoundtrip(TestCase):
+    def roundtrip(self, code):
+        if isinstance(code, str):
+            code = code.encode('utf-8')
+        return untokenize(tokenize(BytesIO(code).readline)).decode('utf-8')
+
+    def test_indentation_semantics_retained(self):
+        """
+        Ensure that although whitespace might be mutated in a roundtrip,
+        the semantic meaning of the indentation remains consistent.
+        """
+        code = "if False:\n\tx=3\n\tx=3\n"
+        codelines = self.roundtrip(code).split('\n')
+        self.assertEqual(codelines[1], codelines[2])
+
+
 __test__ = {"doctests" : doctests, 'decistmt': decistmt}
 
 def test_main():
@@ -1237,6 +1255,7 @@ def test_main():
     support.run_unittest(TestDetectEncoding)
     support.run_unittest(TestTokenize)
     support.run_unittest(UntokenizeTest)
+    support.run_unittest(TestRoundtrip)
 
 if __name__ == "__main__":
     test_main()

Lib/tokenize.py

@@ -244,6 +244,8 @@ class Untokenizer:
 
     def untokenize(self, iterable):
         it = iter(iterable)
+        indents = []
+        startline = False
         for t in it:
             if len(t) == 2:
                 self.compat(t, it)
@@ -254,6 +256,21 @@ class Untokenizer:
                 continue
             if tok_type == ENDMARKER:
                 break
+            if tok_type == INDENT:
+                indents.append(token)
+                continue
+            elif tok_type == DEDENT:
+                indents.pop()
+                self.prev_row, self.prev_col = end
+                continue
+            elif tok_type in (NEWLINE, NL):
+                startline = True
+            elif startline and indents:
+                indent = indents[-1]
+                if start[1] >= len(indent):
+                    self.tokens.append(indent)
+                    self.prev_col = len(indent)
+                startline = False
             self.add_whitespace(start)
             self.tokens.append(token)
             self.prev_row, self.prev_col = end

Misc/NEWS

@@ -60,6 +60,9 @@ Core and Builtins
 Library
 -------
 
+- Issue #20387: Restore semantic round-trip correctness in tokenize/untokenize
+  for tab-indented blocks.
+
 - Issue #24456: Fixed possible buffer over-read in adpcm2lin() and lin2adpcm()
   functions of the audioop module.