codeunit-analyzer（代码单元分析器）

Classic NAV C-AL Performance Analyzer

[!IMPORTANT] This tool targets Classic Microsoft Dynamics NAV (C-AL). It is NOT for Business Central AL. SETLOADFIELDS does not exist in Classic NAV — use the setloadfields sub-command only if you have explicitly migrated to Business Central.

Zero-dependency C-AL analysis tool using only Python standard library. Focuses on Classic NAV performance patterns.

✨ Key Features

• ✅ No external dependencies - Pure Python standard library
• ✅ No virtual environment needed - Works immediately
• ✅ Fast & lightweight - Simple regex-based parsing
• ✅ Windows compatible - No Unicode issues
• ✅ Offline capable - No API server required

⛔ Read-Only Rule — Never Modify Source Files

[!CAUTION] This skill is strictly read-only. It must never write to, overwrite, or modify any codeunit, table, or page source file (.cs, .c-al, or any other AL file). Allowed actions: - Print a report to the console. - Save findings to a separate output/report file (JSON, Markdown, etc.). - If the user explicitly asks to apply a fix, copy the file first (e.g., <name>.patched.cs) and modify only the copy — never the original. If a user asks to "fix" or "update" a codeunit, always confirm whether they want: 1. A report with suggested code changes (default), or 2. A copy of the file with the fix applied. Never silently change source files.

Sub-Commands

1. list - Show All Codeunits

List all available.cs and.c-al files in the workspace.

2. analyze - Deep Codeunit Analysis

Analyze a specific codeunit with bottleneck detection and dependency extraction.

3. scan - Project-Wide Bottleneck Scan

Scan all codeunits for performance issues and generate ranked report.

4. optimize - Optimization Recommendations

Get actionable refactoring suggestions with code examples.

5. setloadfields - SETLOADFIELDS / SELECT-* Audit

Dedicated audit that finds every FINDSET, FINDFIRST, FIND, FINDLAST, or GET without SETLOADFIELDS and ranks them by urgency score — combining:

• Column-width multiplier — wider tables (more fields) waste more network bytes per row.
• Row-count multiplier — large tables amplify the cost of every unnecessary column.
• Loop penalty — reads inside loops are fired N times, so the cost is multiplied.
• FlowField bonus — Business Central calculates *all* FlowFields on a SELECT *; pinning fields with SETLOADFIELDS skips those expensive sub-queries entirely.

The table metadata JSON files can optionally include field_count (int) and has_flow_fields (bool) to enable precise scoring. If absent, the analyzer falls back to neutral multipliers.

Pre-Flight Setup (Run This First, Every Time)

Before any command, resolve two things: the Python executable and the codeunits, tables, or pages directories. Do both in one shot:

# 1. Find Python
PYTHON=$(command -v python3 || command -v python || echo "uv run python")
$PYTHON --version || { echo "Python not found. Install via: brew install python  OR  uv python install 3.12"; exit 1; }

# 2. Find the SKILL path
SKILL="$(dirname "$(dirname "$(find . -path '*codeunit-analyzer/scripts/analyze.py' | head -1)")")"

# 3. Find the codeunits, tables, or pages directories — check common locations
CODEUNITS_DIR=$(
  for candidate in codeunits data/codeunits src/codeunits .agents/data/codeunits; do
    [ -d "$candidate" ] && { echo "$candidate"; break; }
  done
)

# 4. If not found, search from project root
if [ -z "$CODEUNITS_DIR" ]; then
  CODEUNITS_DIR=$(find . -name "*.c-al" -o -name "*.cs" 2>/dev/null | head -1 | xargs dirname 2>/dev/null)
fi

# 5. Export so the script picks it up automatically
export CODEUNITS_DIR
echo "Python : $PYTHON"
echo "Skill  : $SKILL/scripts/analyze.py"
echo "Codeunits: $CODEUNITS_DIR"

If CODEUNITS_DIR is still empty, ask the user where their .cs/.c-al files are and set it manually: export CODEUNITS_DIR=/path/to/codeunits

With those three variables set, all commands become:

$PYTHON $SKILL/scripts/analyze.py list
$PYTHON $SKILL/scripts/analyze.py scan
$PYTHON $SKILL/scripts/analyze.py analyze <file>

Quick Start

No setup required beyond the pre-flight above.

# List all available codeunits
$PYTHON $SKILL/scripts/analyze.py list

# Analyze a specific codeunit
$PYTHON $SKILL/scripts/analyze.py analyze 1.cs

# Scan all codeunits for bottlenecks
$PYTHON $SKILL/scripts/analyze.py scan

# Save scan results to file
$PYTHON $SKILL/scripts/analyze.py scan -o bottlenecks.json

# Get optimization suggestions
$PYTHON $SKILL/scripts/analyze.py optimize 80.cs

# SETLOADFIELDS audit — project-wide (ranked by urgency)
$PYTHON $SKILL/scripts/analyze.py setloadfields

# SETLOADFIELDS audit — single file
$PYTHON $SKILL/scripts/analyze.py setloadfields 80.cs

# SETLOADFIELDS audit — save JSON, show top 10 tables
$PYTHON $SKILL/scripts/analyze.py setloadfields --top 10 -o slf_report.json

Usage Patterns

Pattern 1: List Files

User says:

• "List all codeunits"
• "Show me available files"
• "What codeunits, tables, or pages can I analyze?"

Action:

$PYTHON $SKILL/scripts/analyze.py list

Output: Table showing: File name | Object name | ID

Pattern 2: Analyze Specific Codeunit or Table

User says:

• "Analyze codeunit 80"
• "Explain what codeunit 1.cs does"
• "Analyze table 18"
• "Analyze "

Action:

$PYTHON $SKILL/scripts/analyze.py analyze <filename>

Steps:

1. If no filename provided, run list first
2. Execute analysis command
3. Display bottlenecks sorted by severity
4. Show dependencies and side effects
5. If critical issues found, suggest running optimize

Output includes:

• Codeunit overview (ID, name, procedures)
• Performance bottlenecks with severity levels
• Dependencies (tables referenced)
• Side effects (INSERT/MODIFY/DELETE/COMMIT)

Pattern 3: Scan All Codeunits

User says:

• "Scan for bottlenecks"
• "Find all performance issues"
• "Which codeunits are slow?"
• "Performance audit"

Action:

$PYTHON $SKILL/scripts/analyze.py scan

Save to JSON:

$PYTHON $SKILL/scripts/analyze.py scan -o bottlenecks_$(date +%Y%m%d).json

Steps:

1. Scan all.cs and.c-al files
2. Detect bottlenecks in each
3. Sort by severity and score
4. Group by codeunit
5. Display top 10 offenders
6. Show critical issues first

Output includes:

• Total issue count by severity
• Top 10 worst codeunits (by score)
• Critical issues detailed (first 5)
• Optional JSON export for tracking

Pattern 4: Optimize Codeunit

User says:

• "Optimize codeunit 80"
• "How can I improve performance?"
• "Fix bottlenecks in "
• "Make it faster"

Action:

$PYTHON $SKILL/scripts/analyze.py optimize <filename>

Steps:

1. Analyze the file for bottlenecks
2. Group issues by severity
3. Create phased optimization plan
4. Show code examples for fixes
5. Estimate performance impact

Output includes:

• Phase 1: Critical fixes (do first)
• Phase 2: High priority fixes (do next)
• Phase 3: Medium priority backlog
• Recommendations summary
• Estimated performance gains

Pattern 5: SETLOADFIELDS Audit

User says:

• "Find all missing SETLOADFIELDS"
• "Which tables are we doing SELECT * on?"
• "We have wide tables — are we loading all columns everywhere?"
• "SETLOADFIELDS performance audit"
• "Where are we fetching all fields without filtering?"

Action (project-wide):

python .skills/codeunit-analyzer/scripts/analyze.py setloadfields

Action (single file):

python .skills/codeunit-analyzer/scripts/analyze.py setloadfields <filename>

Flags:

• --top N — show only the N most urgent tables
• --json — output raw JSON
• -o <file> — save findings to a JSON file

Output includes:

• Summary: total missing SETLOADFIELDS, loop occurrences, FlowField tables
• Table ranked by combined urgency score (column-width × row-count × loop penalty)
• Per-read findings with code fix examples

Interactive Flow

Scenario 1: User invokes with just "list"

User: /codeunit-analyzer list
→ Run: $PYTHON $SKILL/scripts/analyze.py list
→ Show table of all codeunits

Scenario 2: User wants to analyze but doesn't specify file

User: /codeunit-analyzer analyze
→ Run list first to show options
→ Ask: "Which file would you like to analyze?"
→ User selects file
→ Run: $PYTHON $SKILL/scripts/analyze.py analyze <selected_file>

Scenario 3: User provides filename directly

User: /codeunit-analyzer analyze 1.cs
→ Run: $PYTHON $SKILL/scripts/analyze.py analyze 1.cs
→ Display results immediately

Scenario 4: Generic request

User: "Find performance issues"
→ Infer command: scan
→ Run: $PYTHON $SKILL/scripts/analyze.py scan

Detected Bottleneck Patterns

The script detects these Classic NAV performance anti-patterns:

[!NOTE] Patterns are grouped by how they manifest in Classic NAV. SETLOADFIELDS (Business Central only) is intentionally excluded from the main scan.

1. N+1 Query (Critical - 150 points)

Database queries inside loops causing massive slowdown.

Example:

// ❌ Bad: Queries database for each iteration
REPEAT
  Customer.GET(SalesLine."Sell-to Customer No.");
  ...
UNTIL SalesLine.NEXT = 0;

// ✅ Good: Query once, process in memory
Customer.GET(SalesHeader."Sell-to Customer No.");
REPEAT
  ...
UNTIL SalesLine.NEXT = 0;

2. Explicit COMMIT (High - 100 points)

Manual transaction commits that break rollback behavior.

Example:

// ❌ Bad: Breaks automatic rollback
IF Customer.INSERT THEN BEGIN
  COMMIT;  // Don't do this!
  MESSAGE('Customer created');
END;

// ✅ Good: Let NAV handle transactions
IF Customer.INSERT THEN BEGIN
  MESSAGE('Customer created');
END;

3. Nested Loops with Writes (High - 120 points)

Database writes inside nested loops.

Example:

// ❌ Bad: Write operation in nested loop
REPEAT
  ItemEntry.SETRANGE("Item No.", SalesLine."No.");
  IF ItemEntry.FINDSET THEN
    REPEAT
      ItemEntry.Processed := TRUE;
      ItemEntry.MODIFY;  // Slow!
    UNTIL ItemEntry.NEXT = 0;
UNTIL SalesLine.NEXT = 0;

// ✅ Good: Use bulk operations
REPEAT
  ItemEntry.SETRANGE("Item No.", SalesLine."No.");
  ItemEntry.MODIFYALL(Processed, TRUE);  // Fast!
UNTIL SalesLine.NEXT = 0;

4. High Write Density (Medium - 60 points)

Too many individual write operations (>10 per codeunit).

Recommendation: Batch operations or use MODIFYALL/DELETEALL.

5. N+1 Queries Inside Loops (Critical)

Too many database calls inside REPEAT...UNTIL blocks — the single biggest Classic NAV performance killer.

6. LOCKTABLE Overuse (Critical / High)

Classic NAV uses page-level locking. LOCKTABLE inside a loop holds the lock for every iteration, blocking all concurrent users.

Example:

// ❌ Bad: lock held for entire loop
REPEAT
  Table.LOCKTABLE;
  Table.MODIFY;
UNTIL Table.NEXT = 0;

// ✅ Good: acquire once
Table.LOCKTABLE;
IF Table.FINDSET THEN
  REPEAT Table.MODIFY; UNTIL Table.NEXT = 0;

7. FlowField Overuse (High)

Each CALCFIELDS issues a separate SQL sub-query. 4+ calls in one procedure without caching is a Classic NAV sign of denormalisation or missing local variables.

Example:

// ❌ Bad: 4 separate sub-queries
Cust.CALCFIELDS(Balance);
Cust.CALCFIELDS('Sales (LCY)');
Cust.CALCFIELDS('Profit (LCY)');
Cust.CALCFIELDS('Balance Due');

// ✅ Good: one batched call + cache in variables
Cust.CALCFIELDS(Balance, 'Sales (LCY)', 'Profit (LCY)', 'Balance Due');
BalanceDue := Cust.'Balance Due';

8. Key / Index Misalignment (High / Medium)

Classic NAV table reads must follow the defined key order. Filtering on a non-leading key field causes a full table scan. The detector flags single-field SETRANGE filters on large tables as potential key misalignments.

Example:

// ❌ Bad: key is 'Document Type, No.' but we skip the first field
Line.SETRANGE('No.', DocNo);  // full scan!

// ✅ Good: lead with the primary key field
Line.SETRANGE('Document Type', DocType);
Line.SETRANGE('No.', DocNo);
IF Line.FINDSET THEN ...

Output Formatting

Console Output

================================================================================
[1] [CRITICAL] - N Plus One Query
================================================================================
Procedure: PostSalesOrder
Score: 150 points

Issue:
  Database query inside loop - causes N+1 query problem

Recommendation:
  Load all records at once using FINDSET, then process in memory

Code Example:
  // Use FINDSET outside loop
  IF Table.FINDSET THEN
    REPEAT
      // Process here
    UNTIL Table.NEXT = 0;

Severity Levels

• [CRITICAL] - Immediate action required (150 points)
• [HIGH] - Fix this sprint (100-120 points)
• [MEDIUM] - Plan for next sprint (60 points)
• [LOW] - Backlog item

JSON Export Format

[
  {
    "pattern": "n_plus_one_query",
    "severity": "critical",
    "score": 150,
    "procedure": "PostSalesOrder",
    "explanation": "Database query inside loop - causes N+1 query problem",
    "recommendation": "Load all records at once using FINDSET",
    "example": "// Use FINDSET outside loop\nIF Table.FINDSET THEN...",
    "codeunit": {
      "file": "80.cs",
      "object_name": "Sales-Post",
      "object_id": "80"
    }
  }
]

Environment Requirements

Minimal Setup ✨

• Python: 3.10+ (standard library only)

Directory Structure

OneTrade/
├── scripts/
│   └── analyze.py          # Standalone script (zero deps)
├── data/
│   └── codeunits/          # Your .cs/.c-al files
└── .skills/
    └── codeunit-analyzer/  # This skill

That's it! No requirements.txt, no pip install, no virtual environment.

Error Handling

Common Errors

1. No codeunits found

Error: No codeunits found.
Expected directory: c:\Users\...\data\codeunits

Solution: Create the data directory and add your.cs or.c-al files.

2. File not found

Error: FileNotFoundError: '80.cs'

Solution: Run $PYTHON $SKILL/scripts/analyze.py list to see available files.

3. No bottlenecks detected

[OK] No bottlenecks detected - this codeunit is already optimized!

Solution: Great news! No action needed.

AL-Perf Integrated Anti-Patterns

The analyzer incorporates advanced source-correlated checks natively adopted from al-perf:

• Missing SetLoadFields *(Business Central only — not Classic NAV)*: Available via the setloadfields sub-command when explicitly targeting BC. Not triggered in the standard scan. Navision defaults to SELECT * across the network, bloating the buffer.
• Unfiltered FindSet (High): Flags FINDSET operations lacking SETRANGE or SETFILTER, severely impacting SQL Server memory caching by pulling entire tables.
• Event Subscriber Hotspots (Critical): Flags [EventSubscriber] procedures containing data-loops or heavy read operations. Because Navision Event Subscribers run globally and synchronously, any blocked loop here degrades the entire ERP execution context.

Troubleshooting

Script won't run

Check Python version:

python --version  # Should be 3.10+

Run from project root:

$PYTHON $SKILL/scripts/analyze.py list

No files listed

Check data directory:

ls data/codeunits/*.cs | wc -l

If empty, add your C-AL files to data/codeunits/.

Permission denied (Linux/Mac)

chmod +x scripts/analyze.py
$PYTHON $SKILL/scripts/analyze.py list

Performance Notes

• List: < 1 second for 2000+ files
• Analyze: < 2 seconds per file
• Scan: Processes all files sequentially (~1-2 sec per file)
• Optimize: < 2 seconds per file

Note: No caching implemented yet - each run re-parses files. This keeps the code simple and dependency-free.

Examples

Example 1: Quick list

python .skills/codeunit-analyzer/scripts/scripts/analyze.py list

Output:

Found 2414 codeunits:

File                 Object Name                                        ID
==================================================================================
1.cs                 ApplicationManagement                              1
80.cs                Sales-Post                                         80
...

Example 2: Analyze specific file

$PYTHON $SKILL/scripts/analyze.py analyze 80.cs

Output:

================================================================================
CODEUNIT: Sales-Post
================================================================================
ID: 80
Procedures: 15
File: 80.cs

================================================================================
PERFORMANCE BOTTLENECKS (5 issues)
================================================================================

Critical: 2
High: 2
Medium: 1

[Detailed bottleneck information follows...]

Example 3: Full scan

$PYTHON $SKILL/scripts/analyze.py scan -o scan_results.json

Output:

================================================================================
BOTTLENECK SCAN SUMMARY
================================================================================
Total Issues: 147
  Critical: 23
  High: 45
  Medium: 67
  Low: 12

================================================================================
TOP OFFENDERS (by total score)
================================================================================

1. Sales-Post (80.cs)
   Issues: 12 | Critical: 3 | High: 5 | Medium: 4
   Total Score: 850 points

[Top 10 worst codeunits listed...]

Full report saved to: scan_results.json

Tips

1. Start with scan - Get overview before drilling down
2. Save scan results - Track progress over time with dated files
3. Focus on critical - [CRITICAL] issues have biggest impact
4. Test after fixes - Verify changes don't break business logic
5. Use JSON export - Integrate with other tools or CI/CD

Limitations

This is a simplified parser using regex. It may not detect:

• Complex nested logic
• Dynamic table references
• Indirect procedure calls
• Some edge case patterns

For production use, consider the full analysis service with Pydantic schemas and comprehensive parsing.

Comparison: Standalone vs Full Service

Use standalone for: Quick scans, CI/CD, no-dependency environments Use full service for: Detailed analysis, AI insights, production workflows

📄 Exporting Reports as PDF or DOCX

When the user asks for a PDF or Word document of the analysis results, use the dedicated community skills instead of building your own export:

Workflow:

1. Run the analyzer and capture the output (use --json or pipe to a file).
2. Hand the structured findings to the docx or pdf skill to format and export.

Trigger phrases that should prompt you to suggest / use these skills:

• "export the report as a PDF"
• "save this as a Word document"
• "give me a.docx of the bottleneck report"
• "I need a PDF of the scan results"

Related Files

• Main Script: analyze.py
• Documentation: .skills/README.md
• Summary: .skills/SUMMARY.md
• Project Docs: CLAUDE.md

Ready to use! No installation, no dependencies, just Python. 🚀

5. Page Navigation Trigger Abuse (High/Critical)

Heavy queries (GET >= 3), nested database writes (MODIFY), or PAGE.RUN calls inside OnAfterGetRecord and similar UI rendering triggers.

Recommendation: Cache records inside Temp tables or run heavy logic exclusively via user-explicit OnAction events.