2504 Design 5ddbb3b8

Technical Design: resolving-the-rest-of-134-completely

Metadata

• Feature: resolving-the-rest-of-134-completely
• Status: REVIEW
• Created: 2026-02-06
• Author: Factory Design Mode

1. Overview

1.1 Summary

Replace 18 scattered rglob calls across 16 files with calls to fs_utils.collect_files(). This is a mechanical refactor — each call site gets the same result set via the shared single-pass traversal. Two call sites need a minor extension to collect_files() for name-based matching (Dockerfiles). Two sites are documented exceptions (rush.py tiny dir, ast_analyzer.py dynamic pattern).

1.2 Goals

• Eliminate redundant directory traversals (Issue #134)
• Standardize file collection through fs_utils.collect_files()
• Improve filtering consistency (replace fragile string-in-path checks with proper path-parts exclusion)

1.3 Non-Goals

• Cross-call caching/memoization of collect_files() results
• Performance benchmarking
• Refactoring duplicated Dockerfile-finding logic between adapters (separate concern)
• Changing any module's behavior or output

2. Architecture

2.1 High-Level Design

Before:                              After:

module_a ──rglob──┐                  module_a ──┐
module_b ──rglob──┤                  module_b ──┤
module_c ──rglob──┤  filesystem      module_c ──┼── fs_utils.collect_files() ── filesystem
module_d ──rglob──┤                  module_d ──┤     (single rglob("*"))
module_e ──rglob──┘                  module_e ──┘
(5 traversals)                       (1 traversal per call site)

Each call site still invokes collect_files() independently (no shared cache), but each invocation does exactly one rglob("*") instead of N separate rglob("*.ext") calls.

2.2 Component Breakdown

2.3 Data Flow

1. Call site determines root directory and desired extensions
2. collect_files(root, extensions=...) walks root once via rglob("*")
3. Files grouped by extension, returned as dict[str, list[Path]]
4. Call site extracts needed extension bucket: result[".py"]
5. Call site applies any remaining post-filters (e.g., underscore exclusion)

3. Detailed Design

3.1 fs_utils Extension

collect_files() needs one addition: a names parameter for Pattern C (Dockerfile matching).

def collect_files(
    root: Path,
    extensions: set[str] | None = None,
    exclude_dirs: set[str] = _DEFAULT_EXCLUDES,
    names: set[str] | None = None,  # NEW: match by filename substring
) -> dict[str, list[Path]]:

When names is provided, files matching any name pattern are included in a "_by_name" bucket regardless of extension. This handles Dockerfile variants (Dockerfile, prod.Dockerfile, Dockerfile.dev).

Alternative considered: Keep Pattern C sites as-is since they already do rglob("*") with inline filtering. The adapters scan project roots which can be large — using collect_files() gives them exclusion of hidden dirs and __pycache__ for free.

Decision: Add names parameter. It's 4 lines of code and serves 3 call sites.

3.2 Migration Patterns

Pattern A (10 sites): Single-extension

# Before:
py_files = sorted(root_dir.rglob("*.py"))

# After:
from zerg.fs_utils import collect_files
grouped = collect_files(root_dir, extensions={".py"})
py_files = grouped.get(".py", [])

Pattern B (4 sites): Multi-extension loop

# Before:
files = []
for ext in ["*.py", "*.js", "*.ts", "*.go", "*.rs"]:
    files.extend(str(f) for f in target.rglob(ext))

# After:
from zerg.fs_utils import collect_files
grouped = collect_files(target, extensions={".py", ".js", ".ts", ".go", ".rs"})
files = [str(f) for ext in grouped for f in grouped[ext]]

Pattern C (3 sites): Name-based matching

# Before (dive_adapter.py / hadolint_adapter.py):
for p in sorted(root.rglob("*")):
    if not p.is_file():
        continue
    if name.endswith(".Dockerfile") or name.startswith("Dockerfile."):
        results.append(p)

# After:
from zerg.fs_utils import collect_files
grouped = collect_files(root, names={"Dockerfile"})
dockerfiles = grouped.get("_by_name", [])

For detector.py, which classifies ALL files:

# Before:
for child in sorted(directory.rglob("*")):
    ...

# After:
grouped = collect_files(directory, extensions=None)
all_files = sorted(f for ext_files in grouped.values() for f in ext_files)

4. Key Decisions

4.1 Add names Parameter vs Keep Pattern C As-Is

Context: 3 call sites use rglob("*") with name-based filtering. They already do a single traversal but lack exclusion of hidden/cache dirs.

Options:

1. Keep as-is — already single-traversal
2. Add names param to collect_files() — standardize + get free exclusion
3. Create separate find_by_name() function

Decision: Option 2 — add names param.

Rationale: Minimal code addition (4 lines), gives Pattern C sites consistent exclusion behavior, avoids proliferating utility functions.

4.2 Document Exceptions vs Force-Migrate

Context: rush.py:323 scans .gsd/ (tiny), ast_analyzer.py:450 takes a dynamic pattern parameter.

Decision: Document as exceptions.

Rationale: rush.py scans ~20 files max. ast_analyzer.py's dynamic pattern doesn't map to collect_files() API without over-engineering. Cost of migration exceeds benefit.

4.3 Post-Filtering Responsibility

Context: Some sites filter by underscore prefix (wiki.py, analyze.py) or by "test" in name (test_cmd.py). Should collect_files() handle this?

Decision: Keep post-filtering at call sites.

Rationale: These filters are domain-specific (e.g., "skip init.py" is an analysis concern, not a filesystem concern). collect_files() handles universal exclusions (hidden dirs, caches). Domain filters stay local.

5. Implementation Plan

5.1 Phase Summary

5.2 File Ownership

5.3 Dependency Graph

Level 1 (Foundation):
  TASK-001: Extend fs_utils.collect_files() with names param

Level 2 (Core — Batch 1, all parallel):
  TASK-002: dependencies.py         ──depends──▶ TASK-001
  TASK-003: validate_commands.py    ──depends──▶ TASK-001
  TASK-004: test_scope.py           ──depends──▶ TASK-001
  TASK-005: analyze.py (3 sites)    ──depends──▶ TASK-001
  TASK-006: wiki.py                 ──depends──▶ TASK-001

Level 2 (Core — Batch 2, all parallel):
  TASK-007: refactor.py             ──depends──▶ TASK-001
  TASK-008: code_fixer.py           ──depends──▶ TASK-001
  TASK-009: test_cmd.py (2 sites)   ──depends──▶ TASK-001
  TASK-010: security_rules.py       ──depends──▶ TASK-001
  TASK-011: build.py                ──depends──▶ TASK-001

Level 2 (Core — Batch 3, all parallel):
  TASK-012: review.py               ──depends──▶ TASK-001
  TASK-013: detector.py             ──depends──▶ TASK-001
  TASK-014: dive_adapter.py         ──depends──▶ TASK-001
  TASK-015: hadolint_adapter.py     ──depends──▶ TASK-001

Level 3 (Verification):
  TASK-016: Full verification       ──depends──▶ TASK-002..015

Note: All Level 2 tasks are independent of each other (no shared files). Batching is for worker count limits only — with enough workers, all 14 can run simultaneously.

6. Risk Assessment

7. Testing Strategy

7.1 Unit Tests

No new unit tests. This is a pure refactor — existing tests validate behavior.

7.2 Integration Tests

Each task runs the full test suite as verification.

7.3 Verification Commands

• Per-task: python -m pytest tests/ -x -q --tb=short (fast fail)
• Final: ruff check zerg/ + grep -r '\.rglob(' zerg/ --include='*.py' to confirm only exceptions remain

8. Parallel Execution Notes

8.1 Safe Parallelization

• TASK-001 (foundation) must complete first
• All 14 migration tasks (TASK-002..015) are fully independent — no shared files
• TASK-016 (verification) runs after all migrations

8.2 Recommended Workers

• Minimum: 1 worker (sequential)
• Optimal: 5 workers (14 tasks / ~3 per worker)
• Maximum: 14 workers (one per migration task)

8.3 Estimated Duration

• Single worker: ~40 min
• With 5 workers: ~15 min
• Speedup: ~2.7x

9. Documented Exceptions

These rglob calls are intentionally NOT migrated:

10. Approval