- Add test_integration_uploads.py with 12 tests for duplicate upload scenarios
- Add test_ref_count.py with 7 tests for ref_count management
- Fix ArtifactDetailResponse to include sha256 and checksum fields
- Fix health check SQL warning by wrapping in text()
- Update tests to use unique content per test run for idempotency
Previously both Python code AND SQL triggers were incrementing/decrementing
ref_count, causing inconsistent values. Now:
- New artifacts start with ref_count=0 (triggers increment on tag creation)
- Tag INSERT/UPDATE/DELETE triggers handle all ref_count changes
- Python code only logs operations, doesn't manipulate ref_count
- Delete endpoints rely on cascade + triggers for ref_count management
Tested: new uploads, duplicates, tag deletion, package deletion (cascade),
project deletion (cascade) - all ref_count values now correct.
- Add DELETE endpoints for tags, packages, and projects with proper ref_count
decrements for all affected artifacts
- Implement atomic ref_count operations using SELECT FOR UPDATE row-level locking
to prevent race conditions
- Add custom storage exceptions (HashComputationError, S3ExistenceCheckError,
S3UploadError) with retry logic for S3 existence checks
- Handle race conditions in upload by locking artifact row before modification
- Add comprehensive logging for all ref_count changes and deduplication events
- Include ref_count in upload response schema
- Grove → Project
- Tree → Package
- Fruit → Artifact
- Graft → Tag
- Cultivate → Upload
- Harvest → Download
Updated across:
- Backend models, schemas, and routes
- Frontend types, API client, and components
- README documentation
- API endpoints now use /project/:project/packages pattern
- Backend: Python 3.12 with FastAPI, SQLAlchemy, boto3
- Frontend: React 18 with TypeScript, Vite build tooling
- Updated Dockerfile for multi-stage Node + Python build
- Updated CI pipeline for Python backend
- Removed old Go code (cmd/, internal/, go.mod, go.sum)
- Updated README with new tech stack documentation
