[PATCH 0/3] sha256 fixes for filter-branch

From: Jeff King <peff@peff.net>
Date: Mar 10, 2021, 17:06 UTC
Message-ID: <YEj8meoPn/g6tCe3@coredump.intra.peff.net>
In-Reply-To: <8735x3i59m.fsf@evledraar.gmail.com>

On Wed, Mar 10, 2021 at 01:04:37PM +0100, Ævar Arnfjörð Bjarmason wrote:

Show 16 quoted lines

> I found a missing spot that wasn't trivial to fix, so sending an E-Mail:
> 
> In git-filter-branch.sh we have:
> 
> _x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
> _x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
> 
> Then later we have a case condition based on matching a SHA-1:
> 
>         $_x40)
>                 echo "Ref '$ref' was rewritten"
>                 if ! git update-ref -m "filter-branch: rewrite" \
> 
> Just deleting that case arm has filter-branch tests passing, so whatever
> it's meant to do it has zero coverage, which explains why it hasn't
> broken with our tests.

It actually does get covered. Dropping that case-arm means we'll fall through to the one below, which does _roughly_ the same thing with a bunch of extra warnings. But none of the tests actually check the error messages, so they don't notice.

Here's a series which fixes it, plus extra tests to notice the distinction.

  [1/3]: t7003: test ref rewriting explicitly
  [2/3]: filter-branch: drop multiple-ancestor warning
  [3/3]: filter-branch: drop $_x40 glob

 git-filter-branch.sh     | 16 ++--------------
 t/t7003-filter-branch.sh | 31 +++++++++++++++++++++++++++++++
 2 files changed, 33 insertions(+), 14 deletions(-)

-Peff