[Duplicity-talk] Fix for corrupted archives with cygwin and python 2.3

[Nick Burch]

Hi All

I recently hit a problem with duplicity running on cygwin producing 
corrupted tar files. I tracked this down to overflowing the file owner 
header field, in the case where cygwin didn't know the owner.

Cygwin was returning a uid of -1 in the case that it didn't know the owner 
of a file (eg owner was a group). ls -l on such a file returned ??????? 
for the owner. duplicity was passing this value of -1 into the formatting 
%07o, which was returing a value that was too large for the field size. 
This was then causing the tar file to be corrupt. (I think python 2.1 
cheated and returned a value of the right size, so it probably wasn't 
noticed before)

This fix is quite simple - detect the -1 value, and substitute for 0. The 
formatting then results in a number of the correct size, and tar correctly 
handles it as an unknown user

Patch below

Nick


--- tarfile.py.orig     2003-08-10 03:17:21.000000000 +0100
+++ tarfile.py  2004-08-02 10:43:35.000000000 +0100
@@ -387,6 +387,14 @@
         """
         # The following code was contributed by Detlef Lannert.
         parts = []
+
+       # If self.uid or self.gid is -1, the tar file will be corrupted.
+       # therefore, we catch this, and chuck in a suitable looking value
+       if self.uid < 0:
+               self.uid = 0
+       if self.gid < 0:
+               self.gid = 0
+       
         for value, fieldsize in (
                 (self.name, 100),
                 ("%07o" % self.mode, 8),