fix: remove additional tags from image URL names

[Mysterio-17]

Description

When deriving an instance name from an image URL, arch aliases (e.g. arm64 for aarch64, amd64 for x86_64) were not being removed even when the image arch matched the native host arch. Only the canonical arch name was stripped.

Changes

• pkg/limatmpl/locator.go: When the image arch is native, iterate over all archKeywords entries and remove any keyword that maps to the same arch, not just the canonical name.
• pkg/limatmpl/locator_test.go: Added test for arch alias deduplication, added the FreeBSD 15.0 aarch64 URL from the issue to TestRead, and made arch-dependent tests use a nativeName field so they pass on any host architecture.

Fixes: #4629

[Mysterio-17]

Hello @unsuman @jandubois , The vz / default.yaml failure is unrelated... all tests pass but cleanup crashes on rootless containerd not running during nerdctl rm -f coredns. Pre-existing flaky CI issue, this PR only touches pkg/limatmpl/locator.go.
Happy to address any feedback.

[jandubois]

Please remove the dummy commit to trigger CI from the branch.

If CI failure looks like it cannot be a result of the PR change, please wait for a maintainer to rerun the tests. Unfortunately some tests are quite flakey, so this is needed more often than we like.

[jandubois]

This test only tests the alternate native arch removal on the aarch64. It would be better to construct a filename with an alias in it to begin with. Something like

alias = runtime.GOARCH
for keyword, arch := range archKeywords {
    if limatype.IsNativeArch(arch) && arch != keyword {
        alias = keyword
        break
    }
}
image := fmt.Sprintf("...-%s-basic", alias)

[jandubois]

These tests seem redundant with the test above? How can they fail when the test above passes?

[Mysterio-17]

These tests seem redundant with the test above? How can they fail when the test above passes?

Yeah... TestRead calls InstNameFromImageURL() internally, so these can't fail independently. I'll drop the nativeName field and the new TestRead entry, keeping only the direct TestInstNameFromImageURL test.

[Mysterio-17]

@jandubois , I've made the changes, rewrote the test to dynamically use the native arch alias so it works on any host, removed the redundant name checks from TestRead, and dropped the dummy CI commit.

[jandubois]

I only see that you dropped the empty commit. I don't see any other changes.

[Mysterio-17]

Hi @jandubois , I've made the changes, Kindly review it

[jandubois]

This version does not always pick an alias; I've given you a better version already in my previous review. Here is a refined variant:

Suggested change

	alias := runtime.GOARCH
	if arch == alias {
	t.Skip("no alias for native arch")
	}
	image := fmt.Sprintf("linux-%s.raw", alias)
	name := limatmpl.InstNameFromImageURL(image, arch)
	var alias string
	for keyword, arch := range archKeywords {
	if limatype.IsNativeArch(arch) && arch != keyword {
	alias = keyword
	break
	}
	}
	if alias == "" {
	t.Skip("no alias for native arch")
	}
	image := fmt.Sprintf("linux-%s-basic.raw", alias)
	name := limatmpl.InstNameFromImageURL(image, limatype.NewArch(runtime.GOARCH))

This performs the test even when runtime.GOARCH is arm or ppc64le, which in your test would just skip.

[jandubois]

I don't know why you are removing this test; it needs to stay.

[jandubois]

I'm only talking about the expectedName test. It existed before your change, and needs to stay. The additional nativeName test you added is redundant with the earlier test and should be removed.

[Mysterio-17]

Hey @jandubois , I've implemented your suggested archKeywords iteration for the alias test so it works on all architectures including arm/ppc64le. Also restored the expectedName assertion in TestRead with nativeName to handle the arch-dependent output on native hosts.

[jandubois]

I'm only talking about the expectedName test. It existed before your change, and needs to stay. The additional nativeName test you added is redundant with the earlier test and should be removed.

[Mysterio-17]

Thanks for the review! I've updated the alias test to use your suggested loop across archKeywords so it works properly on all architectures. I also removed nativeName entirely and restored all the original expectedName values, updating the test to dynamically check for the stripped arch name inside the test assertion block instead.