Re: [RFC/PATCH 12/12] mktag: bring back some of the verify_object() logic

From: Jeff King <peff@peff.net>
Date: Nov 26, 2020, 08:32 UTC
Message-ID: <X79oKaMARatPS83t@coredump.intra.peff.net>
In-Reply-To: <20201126012854.399-13-avarab@gmail.com>

On Thu, Nov 26, 2020 at 02:28:54AM +0100, Ævar Arnfjörð Bjarmason wrote:

Show 9 quoted lines

> When working on this series I saw too late that I'd removed the mktag
> check for validating the object the tag points to. The fsck_tag() code
> doesn't do this because it's meant for the context of fsck, where
> we're validating reachability anyway.
> 
> We'd need to either refactor fsck_tag() so that it can pass us back
> its "tagged_oid" and the "type_from_string_gently()" value it throws
> away to get rid of the re-parsing of stdin here, or just duplicate the
> logic as I'm doing here.

We have yet another tag parser (because of course there's more): the one in parse_tag_buffer() that we use for reading tags. I think your new function here could just be:

  enum object_type real_type;
  struct tag t; /* yuck! another fake object */

  memset(&t, 0, sizeof(t));
  if (parse_tag_buffer(r, &t, buf, len) < 0)
	die("unable to parse");

  real_type = oid_object_info(r, &t->tagged->oid, NULL);
  if (real_type < 0)
	die("tagged object does not exist");
  if (real_type != t->tagged->type)
	die("tagged object's type does not match tag type field");

I almost wonder if we could simply rely on parse_tag_buffer() instead of fsck_tag(), but it is not nearly as picky about finding potential problems (its goal is the opposite: to return something usable if it's at all possible to do so).

The "yuck" above isn't great. We could use pretend_object_file() for that, though it's a bit heavy-weight (it computes the actual oid!). And also it's a weird one-off that we've talked about getting rid of.

-Peff