IronCore Image V2 proposal

[adracus]

Proposal

Our current image specification has evolved a lot since the beginning.
This proposal tries to cleanly describe how we boot and to enhance upon the ideas we have so far.

[maxmoehl]

I'd structure all the media types as $purpose+$format. For a kernel the type would then be:

application/vnd.ironcore.kernel+efi

And a UKI could be:

application/vnd.ironcore.uki+efi

It gets a bit weird for the raw EFI binaries, but maybe:

application/vnd.ironcore.raw+efi

For the initrd its nested so that would then be:

# uncompressed
application/vnd.ironcore.initrd+cpio
# gzip compressed
application/vnd.ironcore.initrd.cpio+gzip
# ...

[adracus]

Regarding the naming with $purpose+$format - I deliberately chose to name it like e.g. application/vnd.ironcore.initrd.cpio (base) / application/vnd.ironcore.initrd.cpio+xz as it closely follows the style used for the container OCI specification ( https://github.com/opencontainers/image-spec/blob/main/media-types.md ):

In that spec, for example tar layers are also written as application/vnd.oci.image.layer.v1.tar and the + suffix is only used to disambiguate within that format, e.g. application/vnd.oci.image.layer.v1.tar+gzip

[maxmoehl]

In that spec, for example tar layers are also written as application/vnd.oci.image.layer.v1.tar and the + suffix is only used to disambiguate within that format, e.g. application/vnd.oci.image.layer.v1.tar+gzip

And for JSON it's always with a +, this seems completely arbitrary to me. Following what's already there is always preferred so I guess we stick with the arbitrary system and try to reasonably model it to our use-case. If someone figures out what the system is, I'd love to learn.

[adracus]

The +json suffix is still only used for disambiguation inside a type: For example, take a look at vnd.oci.image.config.v1+json: The type is a config, but to disambiguate (it could be represented either in e.g. yaml or toml) they use the +json suffix. For other parts where a specific file format is required, like application/vnd.oci.image.layer.v1.tar the .tar says that it must be .tar and disambiguation is just e.g. +xz if compressed.

Another project that publishes OCI artifacts is Helm that also follows the convention I mentioned ( https://helm.sh/blog/helm-oci-mediatypes/ ). I'd recommend sticking with what is already known throughout the container OCI spec and surrounding projects already do.