- Mark 12 tests as skip pending feature implementation (GitHub #686)
- Tests cover sorting, pagination, search/filter, and download features
- Unblocks Phase 2 test suite from proceeding to Phase 3
- Features identified in issue: sorting by timestamp/level/method/URI/status, pagination controls, filtering by text/level, and download functionality
- Mocked `getNotifications` and `checkUpdates` in `Layout.test.tsx`
- Prevents `UND_ERR_INVALID_ARG` errors caused by unmocked `undici` network requests in JSDOM
- Ensures clean test execution for `Layout` and child components
- Implemented tests for the emergency server (Tier 2) to validate health checks, security reset functionality, and independent access.
- Created a comprehensive suite for system settings feature toggles, ensuring proper state management and API call metrics reporting.
- Removed redundant feature toggle tests from the system settings spec to maintain clarity and focus.
- Enhanced test isolation by restoring default feature flag states after each test.
Update CI pipeline to pass the built Docker image to integration tests as a file artifact instead of pulling from a registry.
Adds explicit list of integration tests to build-image job outputs logic
Adds step to export charon:local image to tarball in linux/amd64 architecture
Updates integration jobs to download and load the image artifact
Resolves "invalid reference format" errors when registry tags are missing or invalid
Enables integration testing on PRs that do not push to registry
Removed log masking for image refs to enable debugging
Added whitespace trimming for digest output
Implemented 'docker manifest inspect' gate to fail fast on invalid refs
Switched to printf for safer output logging
Added explicit validation for IMAGE_NAME and DEFAULT_TAG to prevent empty values
Implemented per-tag validation loop to catch empty or malformed tags before build
Added debug step to echo generated tags immediately before build-push-action
Ensures invalid Docker references are caught early with descriptive errors
Rewrote the Emit image outputs step in the build-image job to robustly handle Docker image references.
Replaced fragile grep parsing with a safe while read loop for multiline tags.
Implemented deterministic prioritization: Digest > Matching Tag > First Tag.
Added explicit error handling to fail the build immediately if no valid reference is found, preventing "invalid reference format" errors in downstream integration jobs.
Changed 4 files
Detailed explanation of:
- What behavior changed: Removed the `integration_gate_ok` shell variable from the `pipeline-gate` job.
- Why the change was necessary: The variable was defined but not used, causing `shellcheck` (via `actionlint`) to fail the pre-commit hook.
- Any important side effects or considerations: None; the logic relying on this condition recalculates it inline using GitHub Actions expressions.
- Make container prune run perform deletions by default (workflow_dispatch default now false for dry_run)
- Enhance prune script to estimate candidate and deleted image sizes (Docker Hub best-effort; GHCR manifest fallback)
- Emit machine-readable summary (`prune-summary.env`) and human-readable summary to the workflow run
- Upload logs + summary as artifacts and expose `space_saved` in the run summary
Why:
- Previously the scheduled job used dry-run by default and only logged candidates; this change makes scheduled pruning effective and provides visibility into storage reclaimed.
Impact:
- Runs will now remove eligible images by default (use dry_run=true to test)
- Size calculations are best-effort and may be incomplete if registry APIs do not expose sizes
- Fixed "Emit image outputs" step to always populate image references
- Primary: uses digest from docker/build-push-action when available
- Fallback: extracts image tag from steps.tags when digest unavailable
- Ensures image_ref_dockerhub is never empty after successful build
- Added `if: always()` to all gate jobs (integration, coverage, codecov, pipeline)
- Gates now always execute to evaluate upstream job results
- Prevents cascading skips when jobs intentionally skip or fail
- Properly blocks downstream jobs only when gates actually fail
Pipeline now continues through all stages as designed, blocking only on real failures.
Fixes https://github.com/Wikid82/Charon/actions/runs/21803232380
Integration, E2E, and security jobs were being skipped on PR builds because
they required push_image == 'true'. Since the build succeeded and images were
available, these jobs should run regardless of push policy.
Changed conditions to depend on build success and image availability rather
than registry push status. This allows comprehensive testing on all builds
while still optimizing resource usage where needed.
