bpo-33899: Make tokenize module mirror end-of-file is end-of-line behavior (GH-7891)

Most of the change involves fixing up the test suite, which previously made
the assumption that there wouldn't be a new line if the input didn't end in
one.

Contributed by Ammar Askar.

Lib/test/test_tokenize.py

@@ -1,7 +1,8 @@
 from test import support
 from tokenize import (tokenize, _tokenize, untokenize, NUMBER, NAME, OP,
                      STRING, ENDMARKER, ENCODING, tok_name, detect_encoding,
-                     open as tokenize_open, Untokenizer, generate_tokens)
+                     open as tokenize_open, Untokenizer, generate_tokens,
+                     NEWLINE)
 from io import BytesIO, StringIO
 import unittest
 from unittest import TestCase, mock
@@ -11,27 +12,51 @@ import os
 import token
 
 
+# Converts a source string into a list of textual representation
+# of the tokens such as:
+# `    NAME       'if'          (1, 0) (1, 2)`
+# to make writing tests easier.
+def stringify_tokens_from_source(token_generator, source_string):
+    result = []
+    num_lines = len(source_string.splitlines())
+    missing_trailing_nl = source_string[-1] not in '\r\n'
+
+    for type, token, start, end, line in token_generator:
+        if type == ENDMARKER:
+            break
+        # Ignore the new line on the last line if the input lacks one
+        if missing_trailing_nl and type == NEWLINE and end[0] == num_lines:
+            continue
+        type = tok_name[type]
+        result.append(f"    {type:10} {token!r:13} {start} {end}")
+
+    return result
+
 class TokenizeTest(TestCase):
     # Tests for the tokenize module.
 
     # 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.
+    # code, print out a table with tokens. The ENDMARKER, ENCODING and
+    # final NEWLINE are omitted for brevity.
 
     def check_tokenize(self, s, expected):
         # Format the tokens in s in a table format.
-        # The ENDMARKER is omitted.
-        result = []
+        # The ENDMARKER and final NEWLINE are omitted.
         f = BytesIO(s.encode('utf-8'))
-        for type, token, start, end, line in tokenize(f.readline):
-            if type == ENDMARKER:
-                break
-            type = tok_name[type]
-            result.append(f"    {type:10} {token!r:13} {start} {end}")
+        result = stringify_tokens_from_source(tokenize(f.readline), s)
+
         self.assertEqual(result,
                          ["    ENCODING   'utf-8'       (0, 0) (0, 0)"] +
                          expected.rstrip().splitlines())
 
+    def test_implicit_newline(self):
+        # Make sure that the tokenizer puts in an implicit NEWLINE
+        # when the input lacks a trailing new line.
+        f = BytesIO("x".encode('utf-8'))
+        tokens = list(tokenize(f.readline))
+        self.assertEqual(tokens[-2].type, NEWLINE)
+        self.assertEqual(tokens[-1].type, ENDMARKER)
+
     def test_basic(self):
         self.check_tokenize("1 + 1", """\
     NUMBER     '1'           (1, 0) (1, 1)
@@ -922,14 +947,9 @@ async def f():
 class GenerateTokensTest(TokenizeTest):
     def check_tokenize(self, s, expected):
         # Format the tokens in s in a table format.
-        # The ENDMARKER is omitted.
-        result = []
+        # The ENDMARKER and final NEWLINE are omitted.
         f = StringIO(s)
-        for type, token, start, end, line in generate_tokens(f.readline):
-            if type == ENDMARKER:
-                break
-            type = tok_name[type]
-            result.append(f"    {type:10} {token!r:13} {start} {end}")
+        result = stringify_tokens_from_source(generate_tokens(f.readline), s)
         self.assertEqual(result, expected.rstrip().splitlines())
 
 
@@ -1022,8 +1042,8 @@ class Test_Tokenize(TestCase):
             else:
                 return b''
 
-        # skip the initial encoding token and the end token
-        tokens = list(_tokenize(readline, encoding='utf-8'))[1:-1]
+        # skip the initial encoding token and the end tokens
+        tokens = list(_tokenize(readline, encoding='utf-8'))[1:-2]
         expected_tokens = [(3, '"ЉЊЈЁЂ"', (1, 0), (1, 7), '"ЉЊЈЁЂ"')]
         self.assertEqual(tokens, expected_tokens,
                          "bytes not decoded with encoding")
@@ -1039,8 +1059,8 @@ class Test_Tokenize(TestCase):
             else:
                 return b''
 
-        # skip the end token
-        tokens = list(_tokenize(readline, encoding=None))[:-1]
+        # skip the end tokens
+        tokens = list(_tokenize(readline, encoding=None))[:-2]
         expected_tokens = [(3, '"ЉЊЈЁЂ"', (1, 0), (1, 7), '"ЉЊЈЁЂ"')]
         self.assertEqual(tokens, expected_tokens,
                          "string not tokenized when encoding is None")
@@ -1351,18 +1371,21 @@ class TestTokenize(TestCase):
 
         # Test that 500 consequent, one-line defs is OK
         toks = list(tokenize(BytesIO(buf.encode('utf-8')).readline))
-        self.assertEqual(toks[-2].string, 'OK') # [-1] is always ENDMARKER
+        self.assertEqual(toks[-3].string, 'OK') # [-1] is always ENDMARKER
+                                                # [-2] is always NEWLINE
 
     def assertExactTypeEqual(self, opstr, *optypes):
         tokens = list(tokenize(BytesIO(opstr.encode('utf-8')).readline))
         num_optypes = len(optypes)
-        self.assertEqual(len(tokens), 2 + num_optypes)
+        self.assertEqual(len(tokens), 3 + num_optypes)
         self.assertEqual(tok_name[tokens[0].exact_type],
                          tok_name[ENCODING])
         for i in range(num_optypes):
             self.assertEqual(tok_name[tokens[i + 1].exact_type],
                              tok_name[optypes[i]])
         self.assertEqual(tok_name[tokens[1 + num_optypes].exact_type],
+                         tok_name[token.NEWLINE])
+        self.assertEqual(tok_name[tokens[2 + num_optypes].exact_type],
                          tok_name[token.ENDMARKER])
 
     def test_exact_type(self):
@@ -1515,7 +1538,7 @@ class TestRoundtrip(TestCase):
         self.check_roundtrip("if x == 1:\n"
                              "    print(x)\n")
         self.check_roundtrip("# This is a comment\n"
-                             "# This also")
+                             "# This also\n")
 
         # Some people use different formatting conventions, which makes
         # untokenize a little trickier. Note that this test involves trailing

Lib/tokenize.py

@@ -492,8 +492,15 @@ def _tokenize(readline, encoding):
             # BOM will already have been stripped.
             encoding = "utf-8"
         yield TokenInfo(ENCODING, encoding, (0, 0), (0, 0), '')
+    last_line = b''
+    line = b''
     while True:                                # loop over lines in stream
         try:
+            # We capture the value of the line variable here because
+            # readline uses the empty string '' to signal end of input,
+            # hence `line` itself will always be overwritten at the end
+            # of this loop.
+            last_line = line
             line = readline()
         except StopIteration:
             line = b''
@@ -648,6 +655,9 @@ def _tokenize(readline, encoding):
                            (lnum, pos), (lnum, pos+1), line)
                 pos += 1
 
+    # Add an implicit NEWLINE if the input doesn't end in one
+    if last_line and last_line[-1] not in '\r\n':
+        yield TokenInfo(NEWLINE, '', (lnum - 1, len(last_line)), (lnum - 1, len(last_line) + 1), '')
     for indent in indents[1:]:                 # pop remaining indent levels
         yield TokenInfo(DEDENT, '', (lnum, 0), (lnum, 0), '')
     yield TokenInfo(ENDMARKER, '', (lnum, 0), (lnum, 0), '')

Misc/NEWS.d/next/Library/2018-06-24-01-57-14.bpo-33899.IaOcAr.rst

@@ -0,0 +1,3 @@
+Tokenize module now implicitly emits a NEWLINE when provided with input that
+does not have a trailing new line.  This behavior now matches what the C
+tokenizer does internally.  Contributed by Ammar Askar.