Improve versioning

[balopat]

This is regarding #2775.

• hack/new_version now checks via the Github API whether the current version is released or not - if not then fails
• hack/new_version.sh now takes the new version as an argument for easier testing, e.g, try it out with hack/new_version.sh v1beta15!

next PRs will cater for :

• add a PR check that uses this release logic to ensure that no released version is mutated

[codecov]

Codecov Report

Merging #2798 into master will not change coverage.
The diff coverage is n/a.

[dgageot]

Could you create a separate PR to increase the test timeout?

[dgageot]

I'm not sure I like the usage of an alias for latest. We've been there and it's harder to use. For example, you don't get auto import from the IDE when you write latest.SkaffoldConfig{}

[briandealwis]

Could the release.sh script compare .../latest/config.go — or the generated schema? — since the last tagged release?

The last tagged release is available from git describe --match 'v[0-9]*' --abbrev 0 (source).

[balopat]

I'm not sure I like the usage of an alias for latest. We've been there and it's harder to use. For example, you don't get auto import from the IDE when you write latest.SkaffoldConfig{}

The autoimport is a good point. I think the latest pointer is not a vital thing, I can introduce the Released flag without it. I'll split out the timeout increase!

[balopat]

Could the release.sh script compare .../latest/config.go — or the generated schema? — since the last tagged release?

The last tagged release is available from git describe --match 'v[0-9]*' --abbrev 0 (source).

That could work too, thanks for the suggestion!
Pros: it can't be spoofed + no need for an extra bool flag.
Cons: now the contributor won't have a visual confirmation before trying to add a new field - but we can still add a comment before the config version though in the release script.

[dgageot]

LGTM with a few nits