Convert filename to bytes in getlines.

Most files in linecache are stored with the filename as a bytestring
(py2.x). When calling getlines with a unicode filename there will be
implicit conversions using ascii that fails. This is avoided by an
explicit conversion using the filesystem encoding before calling
getlines.

IPython/utils/ulinecache.py

@@ -1,43 +1,45 b''
 """Wrapper around linecache which decodes files to unicode according to PEP 263.
 
 This is only needed for Python 2 - linecache in Python 3 does the same thing
 itself.
 """
 import functools
 import linecache
+import sys
 
 from IPython.utils import py3compat
 from IPython.utils import openpy
 
 if py3compat.PY3:
     getline = linecache.getline
     
     # getlines has to be looked up at runtime, because doctests monkeypatch it.
     @functools.wraps(linecache.getlines)
     def getlines(filename, module_globals=None):
         return linecache.getlines(filename, module_globals=module_globals)
 
 else:
     def getlines(filename, module_globals=None):
         """Get the lines (as unicode) for a file from the cache.
         Update the cache if it doesn't contain an entry for this file already."""
+        filename = py3compat.cast_bytes(filename, sys.getfilesystemencoding())
         lines = linecache.getlines(filename, module_globals=module_globals)
         
         # The bits we cache ourselves can be unicode.
         if (not lines) or isinstance(lines[0], unicode):
             return lines
         
         readline = openpy._list_readline(lines)
         try:
             encoding, _ = openpy.detect_encoding(readline)
         except SyntaxError:
             encoding = 'ascii'
         return [l.decode(encoding, 'replace') for l in lines]
 
     # This is a straight copy of linecache.getline
     def getline(filename, lineno, module_globals=None):
         lines = getlines(filename, module_globals)
         if 1 <= lineno <= len(lines):
             return lines[lineno-1]
         else:
             return ''