The getxattr() system call retrieves an "extended attribute" on a file.  When you call it you pass in a buffer and a size.  The expected behavior is, if you pass in a buffer that's too small, the function returns -1 and sets errno to ERANGE.

On a ZFS filesystem on Linux, using the "ZFS On Linux" port:
    http://zfsonlinux.org/

getxattr() does not behave this way.  Instead, it fills the buffer with the first buffer-size bytes of data (without a zero terminator).

Python's implementation of getxattr() interprets this as success.  Which means that, the way it's implemented, if you call getxattr() to retrieve a value that's > 128 bytes in length, you only get the first 128 bytes.  (Happily, we already have a regression test that finds this!)

Attached is a patch fixing this behavior.  It checks the return value of getxattr() to see if the buffer was filled to 100%.  If so, it retries with a larger buffer.

Attached is an updated patch, this time for 3.3.  (So I'm guessing we won't get a "review" link.)  It incorporates Benjamin's two comments.

Okay. Hopefully, we can kill that thing one day...

Kill what thing?  The review link?  I love that thing?

Anyway, we did get a review link, hoopla!

No, this extra code you're adding. :)

2013/8/12 Benjamin Peterson <report@bugs.python.org>:
>
> Benjamin Peterson added the comment:
>
> Okay. Hopefully, we can kill that thing one day...
>
> ----------
>
> _______________________________________
> Python tracker <report@bugs.python.org>
> <http://bugs.python.org/issue18694>
> _______________________________________