Make btrfs a weak dependency in ignition

[nullr0ute]

In Fedora IoT we don't intend on supporting btrfs so it would be useful to make the rpm dep for ignition a weak dependency so it can be excluded if not required.

from the spec:

%if 0%{?fedora}
Requires: btrfs-progs
%endif

[jlebon]

https://src.fedoraproject.org/rpms/ignition/pull-request/43

Now, this will automatically also drop out btrfs-progs from FCOS. Do we want to keep it in? I lean towards yes. Some folks like btrfs and it's not large (nor does it bring in new deps AFAICT). (See also discussions in #33)

[dustymabe]

Sounds good to me. I'm good for keeping btrfs-progs in.

[nullr0ute]

Now, this will automatically also drop out btrfs-progs from FCOS. Do we want to keep it in? I lean towards yes. Some folks like btrfs and it's not large (nor does it bring in new deps AFAICT). (See also discussions in #33)

Could FCOS not just include it specifically in their package manifest?

[jlebon]

Could FCOS not just include it specifically in their package manifest?

Yup, that's what we're discussing. :)

[ajeddeloh]

Another thought: we should see about conditionalizing support for it in ignition itself. So if you build without btrfs configs using it will fail validation.

[jlebon]

coreos/fedora-coreos-config#249

[lucab]

we should see about conditionalizing support for it in ignition itself. So if you build without btrfs configs using it will fail validation.

My understanding is that weak dependencies are conditionalized at runtime package install/compose time, not at package build-time.

I fear that build-time conditionals would require building a matrix of binary packages (for all the possible combinations) or a package system that natively support that (e.g. USE flags). So, while a good idea, I don't think it would help here.

Perhaps we can instead enhance Ignition to probe for btrfs-prog at runtime (if a btrfs filesystem was requested by config) before trying to perform. Same for anything else that we want to demote to a Recommends.

EDIT: clarified.

[nullr0ute]

My understanding is that weak dependencies are conditionalized at runtime, not at build-time.

Do you mean install time? In the case of ostree use cases that would be compose time.

[dustymabe]

I fear that build-time conditionals would require building a matrix of binary packages (for all the possible combinations) or a package system that natively support that (e.g. USE flags). So, while a good idea, I don't think it would help here.

I agree. Let's not try to do anything fancy here.

[jlebon]

PR merged; this will be fixed in the next build!