Fix code_generator.py to ensure unique, tutorial-specific output (not identical boilerplate) #123

Problem Statement

The code_generator.py module was introduced with PR #103 which added Stage 2 Semantic Logic Parsing with BuildPlan support. However, the implementation contained a critical bug in the _build_generation_context() method that would cause a NameError when the code attempted to generate projects.

Root Cause

The _build_generation_context() method (lines 1200-1290) had broken if/else logic:

1. Undefined Variables: The variables features, summary, and key_concepts were referenced in lines 1278, 1287-1288 but were never defined in either code path
2. Broken Control Flow: Lines 1252-1257 appeared outside the if block but before the else block, creating impossible-to-execute code
3. Missing Variable Extraction: The summary and key_concepts extraction logic was entirely missing from both the BuildPlan path and the legacy fallback path

Example Error

# Line 1278 - features is never defined
"features": features,  # NameError: name 'features' is not defined

# Lines 1287-1288 - summary and key_concepts are never defined  
"summary": summary,  # NameError: name 'summary' is not defined
"key_concepts": key_concepts,  # NameError: name 'key_concepts' is not defined

Solution Implemented

1. Fixed Control Flow

• Moved summary and key_concepts extraction to the top of the method (lines 1214-1219)
• Properly structured if/else blocks for BuildPlan vs legacy extraction
• Ensured all variables are defined in all code paths

2. BuildPlan Path (lines 1222-1243)

When a BuildPlan is available:

• Extract title, technologies, features, and tutorial_steps from the BuildPlan
• Create a complete extracted_info dict with all required fields
• Log success with BuildPlan title

3. Legacy Fallback Path (lines 1244-1281)

When no BuildPlan is available:

• Extract title from multiple fallback sources
• Extract technologies from ai_analysis or metadata
• Extract features from extracted_info or ai_analysis
• Derive tutorial_steps using existing helper method
• Update extracted_info with all fields

4. Video-Specific Output Enhancement

Added title comment to styles.css generation to ensure video traceability in all generated files.

Test Results

All uniqueness tests now pass:

tests/test_code_generator_unique_output.py::test_different_videos_produce_different_js PASSED
tests/test_code_generator_unique_output.py::test_different_videos_produce_different_css PASSED
tests/test_code_generator_unique_output.py::test_same_video_produces_deterministic_output PASSED
tests/test_code_generator_unique_output.py::test_title_appears_in_all_generated_files PASSED
tests/test_code_generator_unique_output.py::test_fingerprint_is_unique_per_video PASSED
tests/test_code_generator_unique_output.py::test_accent_color_varies PASSED

Impact

Before Fix:

• Code would crash with NameError when attempting to generate any project
• BuildPlan integration from PR feat: Stage 2 Semantic Logic Parsing — structured BuildPlan extraction #103 was completely broken
• No projects could be generated successfully

After Fix:

• Both BuildPlan and legacy extraction paths work correctly
• Each video produces unique, tutorial-specific output
• Fingerprinting ensures different videos get different colors, content, and structure
• Generated code includes video title, steps, concepts, and features

Files Changed

• src/youtube_extension/backend/code_generator.py (lines 1200-1288)
  • Fixed _build_generation_context() method
  • Updated _generate_vanilla_styles_css() to include title parameter
  • Updated caller to pass title to CSS generator

Related

• Introduced in: PR feat: Stage 2 Semantic Logic Parsing — structured BuildPlan extraction #103
• Fixes critical bug from: Stage 2 Semantic Logic Parsing implementation
• Tests: tests/test_code_generator_unique_output.py