Huge Refactor and prepare for odiff library mode

[shreyassanthu77]

Major Changes

🏗️ Build System Overhaul

• Complete rewrite of the build system in build.zig
• Replaced buildOdiff() function with makeOdiff() using new OdiffBuildOptions struct
• Implemented three-tier module architecture:
  • image_mod - Core image data structure and operations
  • io_mod - Image format handling (imports image_mod)
  • Main module - Core diffing logic (imports both image and io)
• Removed complex cross-compilation detection logic
• Consolidated test configuration into single test_files array
• Changed test command from test-all to test throughout
• Updated dynamic linking logic with better target detection
• Added commented-out library building code for future use

📁 Module Structure Reorganization

New directory structure:

src/
├── image/
│   └── image.zig          # NEW: Core Image struct and operations
├── io/
│   ├── io.zig             # SIMPLIFIED: Removed Image struct, now just format handling
│   ├── bmp.zig
│   ├── jpeg.zig
│   ├── png.zig
│   ├── tiff.zig
│   └── webp.zig
├── lib/                   # NEW: Core library modules
│   ├── root.zig           # MOVED: Main library entry point
│   ├── diff.zig           # MOVED: Core diffing algorithm
│   ├── antialiasing.zig   # MOVED: Antialiasing detection
│   ├── color_delta.zig    # MOVED: Color difference calculations
│   ├── rvv.c              # MOVED: RISC-V vector implementation
│   └── vxdiff.asm         # MOVED: AVX assembly implementation
├── cli.zig                # UPDATED: Import changes
├── main.zig               # UPDATED: Import changes
└── test_*.zig             # CONSOLIDATED: See test changes below

🖼️ Image Structure Extraction

Created dedicated image module (src/image/image.zig):

• Extracted the Image struct from src/io/io.zig into its own module
• Contains all image manipulation methods:
  • slice(), deinit(), readRawPixelAtOffset(), readRawPixel()
  • setImgColor(), makeSameAsLayout(), makeWithWhiteOverlay()
• Optimized SIMD-based white overlay generation with vector operations

📥 IO Module Simplification

Simplified src/io/io.zig:

• Now imports Image from @import("image")
• Simplified to focus solely on format detection and loading/saving
• Updated documentation references from loadImageEx to loadImageWithFormat

📚 Library Module Prep

Moved core library files to src/lib/:

• root.zig - Now serves as main library entry point, exports Image, io, diff, color_delta, antialiasing
• diff.zig - Updated imports to use new image module
• antialiasing.zig - Updated imports
• color_delta.zig - No changes (pure logic)
• rvv.c - RISC-V vector implementation
• vxdiff.asm - AVX assembly implementation

🧪 Test Consolidation

Major test file reorganization:

• DELETED: src/test_io_bmp.zig, src/test_io_jpg.zig, src/test_io_png.zig, src/test_io_tiff.zig, src/test_io_webp.zig (544 lines removed)
• CREATED: src/test_io.zig - Consolidated all IO tests into single file (277 lines)
• UPDATED: All remaining test files with new import structure:
  • Changed from @import("root.zig") to @import("odiff")
  • Updated Image type references from io.Image to lib.Image

🔄 Import Updates Across Codebase

Updated import statements:

• src/cli.zig: Changed from @import("diff.zig") to @import("odiff") and lib.diff
• src/main.zig: Changed from @import("odiff_lib") to @import("odiff")
• All IO format files (bmp.zig, jpeg.zig, etc.): Changed Image import from io.zig to image.zig
• All test files: Updated to use new module structure

🚀 CI/CD Workflow Updates

GitHub Actions changes:

• .github/workflows/build.yml: Changed test-all to test in two places
• .github/workflows/riscv-qemu.yml: Updated path trigger from "src/rvv.c" to "src/lib/rvv.c"

📝 Documentation Improvements

Updated docstrings:

• Fixed references from loadImageEx to loadImageWithFormat
• Fixed references from saveImageEx to saveImageWithFormat

Statistics

• Files changed: 28
• Insertions: 541
• Deletions: 834
• Net change: -293 lines (more modular, less duplicated code)

Summary

This refactoring transforms odiff from a monolithic application structure to a modular library-ready architecture. The key benefits are:

1. Clear separation of concerns - Image data, IO operations, and core algorithms are now in separate modules
2. Library preparation - The new structure makes it easier to expose odiff as a reusable library
3. Reduced code duplication - Eliminated duplicate Image struct definitions
4. Simplified build system - More maintainable and easier to understand
5. Consolidated testing - All IO tests now in one place, reducing maintenance overhead

The changes maintain full backward compatibility while setting up the foundation for future library usage.

[dmtrKovalenko]

Thank you for doing this but somehow these changes lead to 4x performance regression

[shreyassanthu77]

Wait what 😅 I will see what's going on