Merge ~jbfzs/dkimpy:re_backtrack into dkimpy:master

[Jonathan Bastien-Filiatrault]

Refactor canonicalization.py strip_trailing_lines to avoid using re for more consistent processing across python versions.

This fixes some really rare and pathological cases.

Thanks !
Jonathan

[Scott Kitterman (kitterman)]

Thanks.

What python versions did you test this with? It needs testing with at least 2.7, 3.7, and one lower python3 version.

review: Needs Information

[Jonathan Bastien-Filiatrault (jbfzs)]

I am using Python 3.6.6. This is the version that had problems and that this patch has been tested on.

[Scott Kitterman (kitterman)]

OK. I'm going to go ahead and merge this and then do some additional testing.

review: Approve

[Jonathan Bastien-Filiatrault (jbfzs)]

Alright, thanks for the quick feedback and have a nice day !

diff --git a/dkim/canonicalization.py b/dkim/canonicalization.py
index bb04b1c..8c9ffc1 100644
--- a/dkim/canonicalization.py
+++ b/dkim/canonicalization.py
@@ -41,15 +41,21 @@ def compress_whitespace(content):
 
 
 def strip_trailing_lines(content):
-    content =  re.sub(b"(\r\n)*$", b"\r\n", content)
-    # Yes, this is horrible, but regex processing changed in python3.7 and it
-    # is the least horrible solution I came up with for python2.7, python3.7,
-    # and python3 << 3.7 combined support.  Better solution welcome.
-    if len(content) >= 4:
-        if (content[len(content)-4:] == b'\r\n\r\n'):
-            content = content[:len(content)-2]
-    return content
+    end = None
+    while content[:end].endswith(b"\r\n"):
+        if end is None:
+            end = -2
+        else:
+            end -= 2
+
+    if end is None:
+        return content + b"\r\n"
+
+    end += 2
+    if end == 0:
+        return content
 
+    return content[:end]
 
 def unfold_header_value(content):
     return re.sub(b"\r\n", b"", content)