cmd/oci-image-tool: validate all refs by default

[runcom]

This is taking care of #271 (comment) (left a TODO/FIXME for later in case).

I'll adapt everything to be same as per #266 later

$ oci-image-tool validate /home/amurdaca/src/github.com/projectatomic/skopeo/busybox-oci2
reference "12.04": OK
reference "3.1": OK
reference "a59906e33509d14c036c8678d687bd4eec81ed7c4b8ce907b888c607f6a1e0e6": OK
reference "latest": OK
reference "notlatest": OK
reference "tag": OK
/home/amurdaca/src/github.com/projectatomic/skopeo/busybox-oci2: OK

/cc @wking @vbatts @philips @stevvooe

Signed-off-by: Antonio Murdaca runcom@redhat.com

[wking]

Maybe listReferences? In #159, the analogous function is refs.Engine.List().

[wking]

On Thu, Sep 08, 2016 at 06:52:45AM -0700, Antonio Murdaca wrote:

-- File Changes --

M cmd/oci-image-tool/validate.go (2)
M image/descriptor.go (24)
M image/image.go (30)

Update the man page too.

[stevvooe]

@runcom If unknown types are encountered, what is the intended behavior?

I ask because we do need to be able to have references to unknown things but still pass validation. This will make it easier to extend in the future.

[wking]

On Thu, Sep 08, 2016 at 12:47:36PM -0700, Stephen Day wrote:

@runcom If unknown types are encountered, what is the intended behavior?

I think we need to error with “not implemented”-level errors.

I ask because we do need to be able to have references to unknown
things but still pass validation. This will make it easier to extend
in the future.

That makes sense. We might want an --ignore-unknown-media-types flag
to avoid erroring out if we hit an unrecognized type. There's some
more discussion of unrecognized types in 1.

What happens if you recognize the ref type
(e.g. application/vnd.oci.image.manifest.list.v1+json) but not an
-ancestor- [edit: descendant] type (e.g. application/vnd.oci.image.layer.zip) and
--ignore-unknown-media-types is set? Do you print your warning and
return zero? Or is the error→warning softening only for the root
refs? We can have a global error→warning softening and still error
for the application/vnd.oci.image.layer.zip case if the specs for our
objects explicitly list allowed descriptor types [2,3].

[stevvooe]

@wking I was asking @runcom. It's okay to let a conversation develop before commenting.

[wking]

On Thu, Sep 08, 2016 at 01:00:21PM -0700, Stephen Day wrote:

It's okay to let a conversation develop before commenting.

Sure. I was just linking in previous unrecognized-type discussion to
try and avoid a fragmented discussion of that issue.

[runcom]

@stevvooe right now, if the media type is unknown the tool errors out. We should at least soften that to just print a warning. This behavior is in the validate function though, I can address this in a follow up and discuss more about errors/warnings if it's ok for you, this PR just makes sure we "find" every reference under refs/ to shorten the command line flags.

[wking]

On Thu, Sep 08, 2016 at 01:58:18PM -0700, Antonio Murdaca wrote:

This behavior is in the validate function though, I can address
this in a follow up and discuss more about errors/warnings if it's
ok for you, this PR just makes sure we "find" every reference under
refs/ to shorten the command line flags.

I'm fine punting unrecognized media type issues to follow up work,
since that is an orthogonal problem (this PR just increases the
likelyhood that we find unrecognized types).

[stevvooe]

@runcom SGTM. Please open an issue.

[runcom]

@stevvooe @vbatts @philips PTAL

[philips]

LGTM.

@s-urbaniak ?

[stevvooe]

LGTM

[s-urbaniak]

iiii? ;-)

[runcom]

lol

[s-urbaniak]

LGTM, with one small nit

[philips]

LGTM