icd-c-code-refactorer: Local LLM-based Automatic ICD Version C Code Refactoring Tool

icd-c-code-refactorer is a tool that uses local large language models to enable automatic conversion of C source code between different versions of Interface Control Documents (ICD). It aims to provide an AI-driven code refactoring solution for embedded systems and avionics software maintenance, addressing pain points such as tedious manual refactoring and high error rates caused by ICD version updates.

Interface Control Documents (ICD) are authoritative specifications for system component communication protocols in high-reliability fields such as embedded systems and avionics, covering message definitions, communication protocols, timing requirements, version information, etc. With system iterations, ICD versions are updated frequently (for function expansion, protocol optimization, bug fixes, and standard upgrades). Manual modification of corresponding C code faces issues like heavy workload, high error rates, difficult verification, and strong knowledge dependency.

The core idea is to use the code understanding and generation capabilities of local LLMs to automatically parse differences between old and new ICDs, analyze existing C code, generate code compliant with the new ICD, and keep business logic unchanged. Advantages of local LLMs include data security (no leakage of sensitive information), offline availability (suitable for isolated environments), cost control (no API fees), and controllable latency. The technical architecture includes an ICD parser (supporting multiple formats), a difference analysis engine (field/message/protocol-level differences), a code analyzer (identifying structures/encoding/decoding/verification logic), and an LLM code generator (prompt engineering, code generation, etc.).

Applicable to fields such as avionics (avionics system upgrades, software reuse, compliance with DO-178C), automotive electronics (vehicle platform adaptation, supplier interface updates, compliance with AUTOSAR), industrial control (system upgrade migration, protocol conversion, compliance with IEC61131-3), and national defense/military industry (weapon system upgrades, compliance with MIL-STD, cross-platform porting).

Supports structure definition updates (adding/adjusting fields), encoding/decoding function refactoring (updating pack/unpack functions), and verification logic updates (adjusting CRC calculation ranges). Local LLM options include CodeLlama, StarCoder, DeepSeek-Coder, Qwen-Coder, etc. Prompt design needs to include elements such as task description, ICD change instructions, original code, and requirements.

The correctness of refactored code is ensured through static analysis (compilation checks, code standards, complexity analysis), unit testing (boundary/random/regression testing), and formal verification (memory safety, functional equivalence, protocol compliance).

Current limitations include insufficient accuracy in handling complex logic, limited global understanding of large codebases, and the need for domain-specific customization. Usage recommendations: manually review generated code, apply incrementally (starting with simple scenarios), and support with comprehensive testing.

In the future, we can expect higher accuracy (in-depth semantic understanding), broader support (multiple languages/domain-specific languages), intelligent interaction (integration with development environments), and end-to-end automation (from requirements to testing). This tool demonstrates AI empowerment in embedded development; although it cannot replace engineers, it can improve efficiency, reduce errors, and allow engineers to focus on creative work.