Upgrade to coverage 3.7.1 and have it auto-build itself on first use.

third_party/coverage-3.7.1/coverage/phystokens.py

Index: third_party/coverage-3.7.1/coverage/phystokens.py
diff --git a/third_party/coverage-3.6/coverage/phystokens.py b/third_party/coverage-3.7.1/coverage/phystokens.py
similarity index 93%
rename from third_party/coverage-3.6/coverage/phystokens.py
rename to third_party/coverage-3.7.1/coverage/phystokens.py
index 166020e1fac0bd4afe41b1ba2d96e80fd54a1769..99b1d5ba0c79771e43338cc8a37ce09e7085d7e2 100644
--- a/third_party/coverage-3.6/coverage/phystokens.py
+++ b/third_party/coverage-3.7.1/coverage/phystokens.py
@@ -1,7 +1,9 @@
 """Better tokenizing for coverage.py."""
 
 import codecs, keyword, re, sys, token, tokenize
-from coverage.backward import StringIO              # pylint: disable=W0622
+from coverage.backward import set                       # pylint: disable=W0622
+from coverage.parser import generate_tokens
+
 
 def phys_tokens(toks):
     """Return all physical tokens, even line continuations.
@@ -18,7 +20,7 @@ def phys_tokens(toks):
     last_ttype = None
     for ttype, ttext, (slineno, scol), (elineno, ecol), ltext in toks:
         if last_lineno != elineno:
-            if last_line and last_line[-2:] == "\\\n":
+            if last_line and last_line.endswith("\\\n"):
                 # We are at the beginning of a new line, and the last line
                 # ended with a backslash.  We probably have to inject a
                 # backslash token into the stream. Unfortunately, there's more
@@ -74,11 +76,11 @@ def source_token_lines(source):
     is indistinguishable from a final line with a newline.
 
     """
-    ws_tokens = [token.INDENT, token.DEDENT, token.NEWLINE, tokenize.NL]
+    ws_tokens = set([token.INDENT, token.DEDENT, token.NEWLINE, tokenize.NL])
     line = []
     col = 0
     source = source.expandtabs(8).replace('\r\n', '\n')
-    tokgen = tokenize.generate_tokens(StringIO(source).readline)
+    tokgen = generate_tokens(source)
     for ttype, ttext, (_, scol), (_, ecol), _ in phys_tokens(tokgen):
         mark_start = True
         for part in re.split('(\n)', ttext):
@@ -122,7 +124,7 @@ def source_encoding(source):
     cookie_re = re.compile(r"coding[:=]\s*([-\w.]+)")
 
     # Do this so the detect_encode code we copied will work.
-    readline = iter(source.splitlines()).next
+    readline = iter(source.splitlines(True)).next
 
     def _get_normal_name(orig_enc):
         """Imitates get_normal_name in tokenizer.c."""
@@ -177,7 +179,9 @@ def source_encoding(source):
             raise SyntaxError("unknown encoding: " + encoding)
 
         if bom_found:
-            if codec.name != 'utf-8':
+            # codecs in 2.3 were raw tuples of functions, assume the best.
+            codec_name = getattr(codec, 'name', encoding)
+            if codec_name != 'utf-8':
                 # This behaviour mimics the Python interpreter
                 raise SyntaxError('encoding problem: utf-8')
             encoding += '-sig'