Handle CRLF and mixed-ending files

pre_commit_hooks/end_of_file_fixer.py

@@ -5,6 +5,10 @@
 from collections.abc import Sequence
 from typing import IO
 
+LF = b'\n'
+CR = b'\r'
+CRLF = b'\r\n'
+
 
 def fix_file(file_obj: IO[bytes]) -> int:
     # Test for newline at end of file
@@ -15,13 +19,20 @@ def fix_file(file_obj: IO[bytes]) -> int:
         return 0
     last_character = file_obj.read(1)
     # last_character will be '' for an empty file
-    if last_character not in {b'\n', b'\r'} and last_character != b'':
+    if last_character not in {LF, CR} and last_character != b'':
+        # Check for consistent CRLF usage
+        file_obj.seek(0, os.SEEK_SET)
+        content = file_obj.read()
+        lf_count = content.count(LF)
+        crlf_count = content.count(CRLF)
+        # Use CRLF only if all line endings are CRLF
+        ending = CRLF if crlf_count > 0 and crlf_count == lf_count else LF
         # Needs this seek for windows, otherwise IOError
         file_obj.seek(0, os.SEEK_END)
-        file_obj.write(b'\n')
+        file_obj.write(ending)
         return 1
 
-    while last_character in {b'\n', b'\r'}:
+    while last_character in {LF, CR}:
         # Deal with the beginning of the file
         if file_obj.tell() == 1:
             # If we've reached the beginning of the file and it is all
@@ -38,7 +49,7 @@ def fix_file(file_obj: IO[bytes]) -> int:
     # newlines.  If we find extraneous newlines, then backtrack and trim them.
     position = file_obj.tell()
     remaining = file_obj.read()
-    for sequence in (b'\n', b'\r\n', b'\r'):
+    for sequence in (LF, CRLF, CR):
         if remaining == sequence:
             return 0
         elif remaining.startswith(sequence):

tests/end_of_file_fixer_test.py

@@ -18,6 +18,8 @@
     (b'foo\n\n\n', 1, b'foo\n'),
     (b'\xe2\x98\x83', 1, b'\xe2\x98\x83\n'),
     (b'foo\r\n', 0, b'foo\r\n'),
+    (b'foo\r\nbar', 1, b'foo\r\nbar\r\n'),
+    (b'foo\nbar\r\nbaz', 1, b'foo\nbar\r\nbaz\n'),
     (b'foo\r\n\r\n\r\n', 1, b'foo\r\n'),
     (b'foo\r', 0, b'foo\r'),
     (b'foo\r\r\r\r', 1, b'foo\r'),