majiayu000 Debugging
Systematic debugging using hypothesis testing and root cause analysis.
When to Use
- Error investigation (exceptions, crashes)
- Unexpected behavior diagnosis
- Performance issue root cause
- Integration/dependency failures
MCP Workflow
# 1. Get code at error location
serena.find_symbol(name_path="ErrorClass/method", include_body=True)
# 2. Trace call chain
serena.find_referencing_symbols(name_path="ErrorClass/method")
# 3. Search for similar patterns
serena.search_for_pattern("null.*check|error.*handling", relative_path="src/")
# 4. Check past similar issues
claude-mem.search(query="<error-type>", project="<project>")
# 5. Framework error handling
context7.get-library-docs("<framework>", topic="exception handling")
Workflow
1. Observe
- Error message and stack trace
- Steps to reproduce
- When did it start? (recent changes?)
- Scope (one user? all?)
2. Hypothesize
Based on: [observation]
Hypothesis: [proposed cause]
Test: [how to verify]
3. Test
serena.find_symbol("SuspectedClass/method", include_body=True)
serena.find_referencing_symbols("SuspectedCause")
# Log analysis, local reproduction
4. Fix
1. Write failing test that reproduces bug
2. Implement minimal fix
3. Verify test passes
4. Run full test suite
5. Prevent
- Add edge case tests
- Improve validation
- Update documentation
Common Patterns
Null/Undefined
Symptoms: NullPointerException, "undefined is not a function" Investigate:
- Find where null originates
- Trace back through call chain
- Check data presence assumptions
Race Condition
Symptoms: Intermittent, works in debug mode Red flags:
- Shared mutable state
- Missing synchronization
- Execution order assumptions
Data Inconsistency
Investigate:
- Trace data flow source to error
- Check transformation logic
- Verify cache invalidation
Investigation Techniques
Git Bisect
git bisect start
git bisect bad
git bisect good <commit>
# Test middle commit, mark good/bad
git bisect reset
Stack Trace Analysis
Start at bottom (entry point)
Move up to error line
Examine code at that line
Check what could be null/invalid
Divide and Conquer
A -> B -> C -> D (output wrong)
Check B: correct? -> problem in C or D
wrong? -> problem in A or B
Output Format
## Bug: [Title]
### Symptoms
- Error: [message]
- Location: `file:line`
### Investigation
**Hypothesis 1:** [description]
- Test: [method]
- Result: Confirmed / Ruled Out
### Root Cause
[Identified cause] at `file:line`
### Fix
[Description]
### Prevention
- [ ] Test added
- [ ] Similar patterns checked
Checklist
- Can reproduce consistently
- Stack trace analyzed
- Root cause identified (not just symptom)
- Failing test written
- Fix implemented
- Full test suite passes