This was using remote.WithTransport to set it manually, and this is much
simpler (also annotates with the go-containerregistry version). This is
really nice because it will give us at least some version information
where we had none before (for non-releases).
Before:
ko
ko/(devel)
ko/v0.7.0
After:
ko go-containerregistry/v0.4.0
ko/(devel) go-containerregistry/v0.4.0
ko/v0.7.0 go-containerregistry/v0.4.0
We dropped this flag, so we're always in non-strict mode by default,
so if there is a ko:// reference that fails to build, we don't error
out. This drops some guards so that we are always in strict mode and
only build ko:// things.
* Move ko binary to root of project.
* go.mod says required golang version is 1.14
* go install uses the folder name
* lint and copyright bump
* moving the test file changed the test hash.
* more refs of cmd/test
* add deprecation message
This was previously `ko auth login`, which I've hidden, and perhaps we
can alias to `ko login` in the future, but in theory folks could have
been using `ko auth get`, and I don't want to break that for no reason.
* Allow comma-separated list of platforms
* Parse platform spec once
* Update --platform docs in README
* Update pkg/build/gobuild_test.go
Co-authored-by: Matt Moore <mattmoor@vmware.com>
* Return err for bad defaulting for --platform
Also respect GOARM as variant if the goarch is arm.
* Refactor platform matching
* Update README.md to mention GOARM
* Fix travis test
Co-authored-by: Matt Moore <mattmoor@vmware.com>
* Add ctx to publish.Interface
I noticed that hitting ctrl-C didn't work when pushing images, this
should fix that.
* Use context everywhere that makes sense
* Allow ko to produce estargz layers.
This pulls in the latest google/go-containerregistry, which enables folks to set `GGCR_EXPERIMENT_ESTARGZ=1` to enable `ko` to start producing layers compatible with the estargz format, which enables the containerd estargz-snapshotter to lazy-load parts of images.
* Add README
This simplifies the logic for creating layers and relies on the new option instead of pre-gzipping (and gunzip being faster). The net effect of this should be the elimination of an extra gunzip, but it should not materially impact the memory pressure because we were ALREADY caching the gzipped layers in-memory, we were just doing it on the caller side, which provided less benefit!
* Refactor the publish.Namer to produce the full repo name.
* Add --naked to have image publishing use ${KO_DOCKER_REPO}:tag
* Change flag name, add disclaimer
