Benjamin Gilbert

> Hmm, right. I will look at this. That said, this is an area where I'd argue we _do_ want some platform specifics, because e.g. `aws` shouldn't have a GRUB...

I assume we should emphasize nftables rather than legacy iptables?

`install` likewise installs FCOS by default (except if there's osmet data, of course). Now that there's stream metadata for RHCOS, we'd ideally support something like `--distribution rhcos --stream 4.9` in...

Thinking about this some more, I don't think it's practical to support RHCOS stream metadata in coreos-installer for now. Even if the user knows that they're running RHCOS 4.8, OCP...

> Mmm, I wouldn't say that's true anymore; we landed openshift/enhancements#679 which does include this information in the cluster. Ah, okay, I didn't know about the configmap. Is there a...

We should make a point of emphasizing this in the Container Linux quickstart docs. Regressions that affect particular environments are often reported only after they reach stable, which implies that...

The alias (and @travier's shell function) is also confusing because it won't work for the `install` subcommand: it doesn't set `--privileged` and the bind mounts for `/dev` and `/run/udev`, and...

coreos-installer may be used to prepare a boot device for a system other than the one being installed. For example, an Ubuntu VM host might use the coreos-installer container to...

We intentionally preserve the original filename, so that the OS version number is included in the name. You can capture the stdout of the `download` command to get the name...

Yeah, makes sense. The other issue is that with `--format pxe`, we download three separate artifacts, and it's not clear how to cleanly let the user specify all three filenames.