Bulk Docstring Addition

Bulk Docstring Addition

When to Use

• User asks to add/improve docstrings across a Python package
• Documentation coverage audit reveals gaps
• Pre-release documentation cleanup

Key Principle

Only add docstrings — never change logic or signatures. This is a pure documentation task.

Step 1: Rough Coverage Scan (optional, for prioritization)

Use grep-based ratio to rank files by docstring coverage. Lower ratio = worse coverage:

cd <package_root>
for f in $(find . -name "*.py" | sort); do
  defs=$(grep -c "^\s*def \|^\s*class " "$f" 2>/dev/null || echo 0)
  docs=$(grep -c '"""' "$f" 2>/dev/null || echo 0)
  if [ "$defs" -gt 0 ]; then
    ratio=$(echo "scale=2; $docs / ($defs * 2)" | bc 2>/dev/null)
    echo "$f | defs=$defs | docs=$docs | ratio=$ratio"
  fi
done | sort -t'=' -k4 -n

Files with ratio < 0.70 need the most attention.

Step 2: Precise Gap Detection with AST

This is the critical step — use Python's ast module for exact identification. Grep ratios are rough; AST parsing is authoritative.

# Run via terminal heredoc (python3 << 'PYEOF' ... PYEOF)
import ast, os

base = '.'
missing = []
for root, dirs, files in os.walk(base):
    for f in files:
        if f.endswith('.py'):
            fp = os.path.join(root, f)
            try:
                with open(fp) as fh:
                    tree = ast.parse(fh.read())
            except:
                continue
            for node in ast.walk(tree):
                if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
                    name = node.name
                    # Skip private methods (keep dunder methods like __init__)
                    if name.startswith('_') and not name.startswith('__'):
                        continue
                    docstring = ast.get_docstring(node)
                    if not docstring:
                        kind = 'class' if isinstance(node, ast.ClassDef) else 'def'
                        missing.append((fp, node.lineno, kind, name))

print(f'Total public defs/classes missing docstrings: {len(missing)}')
for fp, line, kind, name in missing:
    print(f'{fp}:{line} {kind} {name}')

Step 3: Read Files and Apply Docstrings

Strategy

1. Read each file that has missing docstrings (use read_file)
2. Batch by file — use multi-file patch mode when possible (V4A patches)
3. Use Google-style format with Args, Returns, Raises, Examples sections

Google-Style Docstring Format

def function_name(arg1: str, arg2: int = 0) -> dict:
    """One-line summary of what this function does.

    Extended description if needed — explain the algorithm,
    context, or non-obvious behavior.

    Args:
        arg1: Description of arg1.
        arg2: Description of arg2.

    Returns:
        Dictionary with keys ``key1`` and ``key2`` containing
        the computed results.

    Raises:
        ValueError: If arg1 is empty.

    Examples:
        >>> result = function_name("hello", 42)
        >>> print(result["key1"])
    """

Patching Approach

• Use patch tool in replace mode for targeted insertions
• For __init__ methods: add docstring right after def __init__(...):
• For inner Config classes (Pydantic): one-line docstring is sufficient
• For inner functions (closures): brief one-liner is fine
• Always verify the patch tool reports lint: ok

Step 4: Verify Zero Gaps Remain

Re-run the AST scanner from Step 2. Target: Total public defs/classes missing docstrings: 0

Scaling: Parallel Subagents for Large Packages

When total file count across target packages exceeds ~50, use delegate_task with parallel subagents:

1. Pre-triage first — Quick coverage scan (Step 1) to identify which packages actually need work. Skip packages already at 100% coverage (e.g. a 2-file package with full docstrings).
2. One subagent per package — Up to 3 in parallel via delegate_task(tasks=[...]). Each subagent gets an isolated context with clear instructions.
3. Scope by gap size — Subagents hit iteration limits (~50 tool calls). A package with 165 files and 43 needing docstrings will only complete ~19 files before exhaustion. For large packages, tell the subagent to prioritize files with ZERO docstrings first, then lowest coverage ratio.
4. Commit per package — After subagents return, commit each package separately for clean git history.

Subagent prompt template:

goal: Add Google-style docstrings to all public functions/classes in <pkg>/
context: Path is <full_path>. <N> files, <M> defs, <D> existing docstring markers.
         Focus on files MISSING docstrings — skip files with good coverage.
         Do NOT modify logic/imports/signatures. Only add docstrings.
toolsets: ["terminal", "file"]

Lesson: Iteration budget planning

• Reading a file + patching it = ~2-3 tool calls per file
• A subagent with 50 iterations can process ~15-20 files with docstrings
• For packages with 40+ files needing work, create a follow-up issue for the remainder

Pitfalls

1. grep ratios mislead — A file can have ratio=0.84 but still have missing docstrings (e.g. module-level docstrings inflate the count). Always use AST for truth.
2. python3 -c with quotes — Complex Python one-liners with nested quotes get blocked by terminal. Use heredoc (python3 << 'PYEOF') or uv run python << 'PYEOF' instead. Some sandboxes block python -c with semicolons entirely.
3. Duplicate files — Packages often have copied/mirrored modules (e.g. marine_analysis/ and marine_engineering/ with identical catenary solvers). Don't forget the copies.
4. Pydantic inner Config classes — These are real classes that AST detects. A one-line """Pydantic configuration for ModelName.""" suffices.
5. Closure/inner functions — AST detects these too (e.g. def run_spectrum(): inside a method). One-liner docstrings are appropriate.
6. Non-unique match in patch — When multiple class Config: or similar exist in one file, use replace mode with enough surrounding context to disambiguate, rather than V4A patch mode.
7. Empty init.py files — These count as files but just need a one-line module docstring. Subagents can bulk-write these with write_file instead of patching.