Am 08.04.2013 23:05, schrieb Jeff King:
> On Mon, Apr 08, 2013 at 02:36:05PM -0400, BJ Hargrave wrote:
>
>> Git 1.8.2.1 includes commit bd54cf17 - archive: handle commits with an
>> empty tree
>>
>> Test 2 of t5004-archive-corner-cases, "tar archive of empty tree is
>> empty", fails on Mac OS X 10.8.3 (with XCode 4.6.1) since the tar
>> command exits with return code 1 on the generated tar file.
>
> Hmm. So I guess the question is: do we need to work around this in the
> test for platforms that do not like empty tar files, or are the empty
> tarfiles we are making wrong somehow?
>
>> tar --version
>> bsdtar 2.8.3 - libarchive 2.8.3
>>
>> It appears that bsdtar does not like the empty tar files created by
>> git archive. An empty tar file created by bsdtar is accepted.
>>
>> tar cT /dev/null | tar t; echo $?
>> 0
>
> That makes me think the latter ("we are wrong"). I don't have my OS X
> box handy; can you provide a sample empty tarfile that it creates?
libarchive (on which bsdtar is based) doesn't like extended pax headers
at the end of archives. Here's the relevant source file:
https://github.com/libarchive/libarchive/blob/master/libarchive/archive_read_support_format_tar.c
tar_read_header() calls header_pax_global() to handle a global pax
header, which in turn calls tar_read_header() again to fetch the next
header. If it reaches the end of the archive then err is set to
ARCHIVE_EOF and "Damaged tar archive" is reported at the end of this
function.
I tried come up with a small patch that convinces it to ignore such a
condition, but it's apparently not as easy as it looks -- I just made
bsdtar report even more obscure errors. Will look deeper into it
later this week.
René
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html