stacklok / toolhive / 24658066489

66%

Build Type

push

github

Committed by web-flow

Commit Message

Fix skill install for qualified namespace/name references and missing parent directory (#4841)

* Fall back to registry lookup when Install receives a qualified namespace/name

* Create parent directory before acquiring lock in WriteFiles

* Block multi-segment OCI refs from triggering registry fallback

Extend the fallback guard in Install and GetContent to also reject
refs with more than one '/' (e.g. ghcr.io/org/skill). Previously only
':' and '@' were checked, so a tagless multi-segment OCI reference
could silently trigger a registry lookup on pull failure.

Add a Debug log line before the fallback so operators can see when
the fallback path is taken.

Fix installFromResolvedRegistry passing opts.Name (the OCI reference
string) to installAndRegister instead of result.Skill.Metadata.Name,
which caused incorrect group entries and broken rollback.

* Add WriteFiles test for missing parent directory

* Extend OCI fallback tests with digest, multi-segment, and ambiguity cases

Add three new test cases to TestInstallQualifiedNameOCIFallback:
- digest ref (@sha256:...) must not trigger registry fallback
- multi-segment OCI ref (ghcr.io/org/skill) must not trigger fallback
- multiple registry matches must surface a conflict error to the caller

Also pin the Pull mock's third argument in the three qualifying-name
cases to the expected reference string instead of gomock.Any().

* Extract isUnambiguousOCIRef helper for registry-fallback guard

Replace the inlined string-based guard in Install and GetContent with a
named helper that combines parsed-form inspection (name.Digest type
assertion) with string checks for the tag and segment-count cases.

A pure parsed-form check is not sufficient: nameref.ParseReference
normalizes "foo/bar" to "foo/bar:latest" (name.Tag), so the parsed
reference cannot distinguish a user-supplied tag from the default. The
helper documents this constraint and combines both signals.

Coverage Stats

42 of 50 new or added lines in 2 files covered. (84.0%)

43 existing lines in 6 files now uncovered.

57962 of 87430 relevant lines covered (66.3%)

62.11 hits per line