Added a shared isolated Ruff preview pass for PLC2701 and wired it into every Python Nx lint target.
Excluded tests, where private helper imports are intentional, from this production-boundary check.
Kept vxdata.sdk.namespaces._SPECIAL_NAMESPACES private and added one narrow # noqa: PLC2701 to the namespace generator.
Preserved legacy Pixi lockfiles by wiring three projects through Nx and Lefthook instead of changing task-bearing manifests that would force lock upgrades.
Verified all ten Python lint targets, Import Linter contracts, lock guards, module boundaries, and a negative test that catches the suppressed import when the suppression is removed.
Folded the initial work into chore: enforce Python import boundaries and later appended focused orchestration commits.
Added advisory ast-grep to the authoritative Python lint flow, pinned the CLI for CI, and opened draft linting PR virdx/mono#353.
Moved complete Python lint behavior into Pixi manifests; Nx targets now only call each project’s frozen Pixi lint task and provide cache metadata.
Upgraded three legacy Pixi v6 locks to v7 in an isolated generated-only commit because manifest edits make the lock guard require the current schema.
Delegated the 43 mechanical ast-grep fixes to a separate worktree and opened virdx/mono#354; it leaves only advisory warnings and hints.
Lessons Learned: Pitfalls
Ruff preview mode changes behavior for already-selected stable rules. Enabling it in the normal project configuration surfaced unrelated legacy findings even with explicit preview rules. A separate --isolated --preview --select PLC2701 pass avoids changing the existing baseline.
Running affected typecheck, test, and build targets concurrently is unsafe because several Nx targets share and delete package output/ and dist/ directories. This caused false build failures, pytest import mismatches, and a temporary missing workspace module. Sequential reruns passed.
The IPython environment set MPLBACKEND=module://matplotlib_inline.backend_inline, which breaks matplotlib tests outside notebooks. External project validation should remove that inherited variable.
Task-only changes in three older Pixi manifests made their legacy lockfiles appear stale and would have forced noisy lock upgrades.
Lessons Learned: Improvements
Document that preview-only Ruff rules should run in an isolated second pass unless the repository has already adopted Ruff preview behavior globally.
Document that build-dependent Nx validation must run sequentially when targets share generated output directories.
Add MPLBACKEND sanitization to native project-command execution guidance for matplotlib projects.
Keep Pixi manifests as the source of truth for Python project tasks. Nx should call those tasks and only add orchestration/cache metadata.
Isolate unavoidable generated lock schema upgrades in their own commit so reviewers can separate them from behavioral changes.
