Now that InitKube, AddRepo, and UpdateDependencies are initialized with
NewSTEPNAME functions, the helper functions in plan.go are
unnecessary--they do too little to be a useful abstraction, and they
aren't complex or frequently-used enough to be worth extracting.
This seems to be be a more natural separation of concerns--the knowledge
of which config fields map to which parts of a Step belong to the Step,
not to the Plan.
I'd like to be able to make calls like NewUpgrade(cfg) rather than
Upgrade{...}.Prepare, but I wouldn't be able to define a NewUpgrade
function while Config is in the helm package; there would be a circular
import when Plan tried to import run.
While writing docs in the previous commit, I noticed that we'd been
inconsistent in the naming scheme. Wikipedia's back-compat article
redirects from "backwards" to "backward", so I figure that's a
reasonable source of authority for which form to use.
The goal with these changes was to give users a clearer, more readable
interface, so we should present that interface up front and only
document the aliases as a backward-compatibility option.
I've renamed the envconfig tags to reflect the switch, but I left the
actual field names the same. I think they're sufficiently meaningful
inside the code, and leaving them unchanged avoids making a bunch of
churn in the rest of the code.
This includes a refactor to the way aliases are processed. I had been
thinking in terms of locking down the aliases names pretty tightly, in
order to provide an error if there are conflicts. After discussion with
@josmo, though, it seems like we can do it the same way we do for
"PLUGIN_"/non-prefixed variables, i.e. quietly override them.
I don't love the mismatch between the helm.Config field (CleanupOnFail)
and the setting name (cleanup_failed_upgrade). I do think the setting
name should contain "upgrade" since it's specific to the upgrade command,
but if I make the config field CleanupFailedUpgrade, it becomes the new
longest field name, and gofmt creates a bunch of churn. Is that a good
enough reason...?
Helm2's --timeout took a number of seconds, rather than the
ParseDuration-compatible string that helm3 uses. For backward-
compatibility, update a bare number into a duration string.
These comments were a reasonable attempt at ensuring the documentation
matched reality, but the checkbox in the pull request template is much
more likely to produce results.
It's a little tricky to find a balance between "brittle" and "thorough"
in this test--I'd like to verify that e.g. the certificate is in
clusters[0].cluster.certificate-authority-data, not at the root. On the
other hand, we can't actually show that it's a valid kubeconfig file
without actually *using* it, so there's a hard upper limit on the
strength of the assertions. I've settled on verifying that all the
settings make it into the file and the file is syntactically-valid yaml.
