Architecture
How gitresolve handles conflicts without a centralized brain.
Core Principles
gitresolve is built on the principle of Local Determinism. Every resolution is produced by a fixed pipeline of rule-based engines that require zero network access and provide bit-identical results for the same input.
AST Integration
Instead of line diffs, we see syntax trees. This allows safe merging of unordered blocks like Go imports or Java annotations.
Rooted IO Sandbox
Mandatory os.Root sandboxing for all file operations (CWE-22 mitigation). IO is mathematically restricted to the repository root.
Pre-write Validation
No resolution is written to disk unless it passes the language parser. Syntax errors trigger immediate manual escalation.
Audit Persistence
Every decision is recorded in a namespaced log, making it possible to query 'why' every resolution occurred months later.
The Resolution Pipeline
Symmetric Marker Identification
Scanning for conflict markers with active brace-balance recovery. If markers are malformed or nested improperly, we escalate immediately.
Heuristic Classification
The engine categorizes the block: TypeIdentical, TypeWhitespace, TypeImport, TypeStructured, or TypeScalar.
Policy-Injected Routing
Active Policy Profiles (strict/aggressive) modify the confidence threshold required to proceed with automation.
Semantic Strategy Execution
Running the actual merge logic. For structured files, this is a deep recursive map merge. For code, it's an AST transformation.
Post-Resolution Syntax Gate
The final check. We run the native language compiler/parser on the 'merged' result. If valid, we write. If invalid, we roll back.
Engine Deep Dive
Visual breakdown of the deterministic resolution engine: from marker identification to post-resolution syntax validation.
Structured Data Engine
For JSON, YAML, and TOML, gitresolve bypasses line-based text merging entirely. It parses both sides into memory, performs a 3-way recursive object merge including conservative array unioning, and re-serializes the result. This prevents common errors where merging two objects results in an invalid JSON comma or duplicated keys.