When tarfile is directed to create tar.gz compressed archive file in a
path different from current, it saves full path information in .gz
header where only filename is required.

This causes problems with decompression utilities, such as 7zip. The
testsuite with patch are attached.

{{{
tar -czf dist\create_tar.tar.gz package
7z l dist\create_tar.tar.gz > tar.out
python test_create.tar.gz.py
7z l dist\create_py.tar.gz > py.out
diff -pu3 tar.out py.out
}}}

{{{
--- tar.out     Fri Dec 26 15:12:42 2008
+++ py.out      Fri Dec 26 15:12:42 2008
@@ -1,10 +1,10 @@

 7-Zip 4.57  Copyright (c) 1999-2007 Igor Pavlov  2007-12-06

-Listing archive: dist\create_tar.tar.gz
+Listing archive: dist\create_py.tar.gz

    Date      Time    Attr         Size   Compressed  Name
 ------------------- ----- ------------ ------------ 
------------------------
-2008-12-26 15:12:41              10240          170  create_tar.tar
+2008-12-26 15:03:39              10240          141  dist/create_py.tar
 ------------------- ----- ------------ ------------ 
------------------------
-                                 10240          170  1 files, 0 folders
+                                 10240          141  1 files, 0 folders
}}}

See also issue 1886 and msg61515 in particular

Lars, what do you think?

Anatoly is right, the gzip file format specification (RFC 1952) says
that the FNAME header field must be the basename of the original
filename. So, this behaviour is not tarfile's fault but that of the gzip
module and should be fixed there.
7zip can still decompress these files, right?

7zip can decompress both, but it still creates "dist/" directory when
decompressing file that is made with Python.

I've noticed this bug with extra path component is actual with "tar" +
"gzip" under windows. If they are executed separately and windows path
with backslashes is used - directory prefix is not stripped. I.e. this
creates archive with invalid header:

{{{
tar -cf dist\create_tar_sep.tar package
gzip -f9 dist\create_tar_sep.tar
}}}

This command is ok:

{{{
tar -cf dist\create_tar_sep.tar package
gzip -f9 dist/create_tar_sep.tar
}}}

For MSYS gzip added a bugreport here:
https://sourceforge.net/tracker2/index.php?func=detail&aid=2474481&group_id=2435&atid=102435

I attach patch for Python 2.6 gzip
I clarified the meaning of self.name to be the basename corresponding to
FNAME field in GZIP file header.

There is a trace of deprecated gzip.filename API - I haven't found any
references to it in documentation, so I removed it. In Python 2.5 it
seemed to mean just filename in read mode and filename + .gz in write
mode even if opened filename did not end with .gz

If FNAME field from gzip header is ignored in read mode, so we want to
make self.filename or self.name available via API - we need to agree
what it should be - basename of archived file or path filename of
archive itself.

I attach for Python 2.5 as well. People will use gzip module for a long
time to build packages and patch will help them to get correct archives.

No further bug fixes are accepted for 2.5 (unless they fix security
problems), so I reject the 2.5 patch.

Lars, is this still accurate ?

The latest patch (4750.gzip.basename.fix.diff) cannot be used the way it
is. The problem is that it uses the name attribute to store the basename
with the .gz extension stripped. This breaks compatibility.

I fixed it in r75935 and r75937.