Carlo Marcelo Arenas Belón <carenas@xxxxxxxxx> writes:
> ea2d20d4c2 ("t5004: avoid using tar for checking emptiness of archive",
> 2013-05-09), introduced a fake empty tar archive to allow for portable
> tests of emptiness without having to invoke tar
>
> 4318094047 ("archive: don't add empty directories to archives", 2017-09-13)
> changed the expected result for its tests from one containing an empty
> directory to a plain empty archive but the portable test wasn't updated
> resulting on them failing again in (at least) NetBSD and OpenBSD
>
> Signed-off-by: Carlo Marcelo Arenas Belón <carenas@xxxxxxxxx>
> ---
> t/t5004-archive-corner-cases.sh | 17 ++++++++---------
> 1 file changed, 8 insertions(+), 9 deletions(-)
>
> diff --git a/t/t5004-archive-corner-cases.sh b/t/t5004-archive-corner-cases.sh
> index ced44355ca..271eb5a1fd 100755
> --- a/t/t5004-archive-corner-cases.sh
> +++ b/t/t5004-archive-corner-cases.sh
> @@ -3,8 +3,12 @@
> test_description='test corner cases of git-archive'
> . ./test-lib.sh
>
> -test_expect_success 'create commit with empty tree' '
> - git commit --allow-empty -m foo
> +# the 10knuls.tar file is used to test for an empty git generated tar
> +# without having to invoke tar because an otherwise valid empty GNU tar
> +# will be considered broken by {Open,Net}BSD tar
> +test_expect_success 'create commit with empty tree and fake empty tar' '
> + git commit --allow-empty -m foo &&
> + perl -e "print \"\\0\" x 10240" >10knuls.tar
> '
OK, so you moved the creation of the file with block of NULs to the
set-up phase of the entire script.
> # Make a dir and clean it up afterwards
> @@ -47,7 +51,6 @@ test_expect_success HEADER_ONLY_TAR_OK 'tar archive of commit with empty tree' '
>
> test_expect_success 'tar archive of empty tree is empty' '
> git archive --format=tar HEAD: >empty.tar &&
> - perl -e "print \"\\0\" x 10240" >10knuls.tar &&
> test_cmp_bin 10knuls.tar empty.tar
> '
And because of that, this one that was added for ea2d20d4 ("t5004:
avoid using tar for checking emptiness of archive", 2013-05-09) is
now simplified. It can use the one that was created in the set-up
phase.
> @@ -106,16 +109,12 @@ test_expect_success 'create a commit with an empty subtree' '
>
> test_expect_success 'archive empty subtree with no pathspec' '
> git archive --format=tar $root_tree >subtree-all.tar &&
> - make_dir extract &&
> - "$TAR" xf subtree-all.tar -C extract &&
> - check_dir extract
> + test_cmp_bin 10knuls.tar subtree-all.tar
> '
And then we avoid the test that assumes that an empty tar archive
can safely and portably extracted, and instead check the emptiness
the same way as the earlier test here ...
> test_expect_success 'archive empty subtree by direct pathspec' '
> git archive --format=tar $root_tree -- sub >subtree-path.tar &&
> - make_dir extract &&
> - "$TAR" xf subtree-path.tar -C extract &&
> - check_dir extract
> + test_cmp_bin 10knuls.tar subtree-path.tar
> '
... and here, too.
OK, and the result is consistent with the "We can help GNU and BSD
tar, but NetBSD tar cannot be salvageable" approach, laid out in the
earlier ea2d20d4 ("t5004: avoid using tar for checking emptiness of
archive", 2013-05-09). Makes sense.
Thanks.
