SymNebula — 符号星云

Scientific Computing Runtime Platform

一张活的、可自动重算的数学草稿纸。 一个可嵌入、可观察、可跨语言调用的确定性数学内核平台。

SymNebula 是一个以 Rust 编写的确定性数学 Runtime Platform，提供 Formula Parser → AST → IR → ConstraintGraph 的全链路公式推演，支持 C ABI / Python Bridge / Event Stream 多层嵌入接口，并通过 Plugin System 实现求解器、节点类型、外部模块的动态扩展。Phase 9 新增 Tensor Runtime、PDE 求解、DSL、编译器、GPU Kernel Fusion、Sparse 矩阵等科学计算能力。

---

架构总览

┌──────────────────────────────────────────────────────────────────┐
│                        Scientific Computing Runtime              │
├──────────────────────────────────────────────────────────────────┤
│  DSL Layer          dsl/ grammar → parser → tensor/pde cmd       │
│  Compiler Layer     compiler/ lowering → GPU/SIMD/JIT            │
│  PDE Runtime        pde/ heat / wave / poisson / NS / FEM        │
│  Tensor Runtime     tensor/ N-D ops → autodiff → solver          │
│  Sparse Runtime     sparse/ CSR/COO → Jacobian → solver          │
│  Graph Optimizer    optimizer/ folding → fusion → partition      │
│  Adaptive Scheduler scheduler/ GPU/CPU/Remote load balance       │
│  GPU Kernel Fusion  gpu/ WGSL kernel fusion                      │
├──────────────────────────────────────────────────────────────────┤
│  Clients (Python / C / AI / Web)                                 │
│     ┌───────┴───────┐                                            │
│     │  External API  │  session/external_api.rs                   │
│     │  C ABI (ffi/) │  extern "C"                                │
│     │  Python Bridge │  pyo3 #[pyclass]                          │
│     └───────┬───────┘                                            │
│  ┌──────────┴──────────┐                                         │
│  │  RuntimeSession      │  session/session.rs + api.rs            │
│  │  add_formula / tick  │  solve / get_value                     │
│  │  snapshot            │  register_solver / solve_with          │
│  └──────────┬──────────┘                                         │
│  ┌──────────┼──────────┐  ┌───────────────┐  ┌─────────────────┐│
│  │Formula   │Constraint│  │ Event Stream  │  │  Observatory    ││
│  │Parser    │Graph     │  │ tokio broadcast│  │  Graph/Sparse/  ││
│  │Token→IR  │DirtyProp │  └───────────────┘  │  Tensor/PDE/... ││
│  └──────────┴──────────┘                     └─────────────────┘│
├──────────────────────────────────────────────────────────────────┤
│  Distributed OS: Cluster / Scheduler / Fabric / Network          │
│  Storage / PersistentWorld / WASM / Monitor                      │
└──────────────────────────────────────────────────────────────────┘

---

里程碑

Phase	主题	状态
Phase 6	Runtime Platform Core (Parser/Graph/Session/CABI/Python)	✅ 已发布
Phase 7	Plugin Architecture (Solver/Node/Module/Sandbox)	✅ 已发布
Phase 8	Distributed Runtime OS (Cluster/Scheduler/Fabric/PersistentWorld)	✅ 已发布
Phase 9	Scientific Computing (Tensor/PDE/Compiler/DSL/Sparse/Optimizer/GPU)	✅ 已发布
Phase 10	Runtime Intelligence (Observability/Monitor/Optimizer/Scheduler)	✅ 已发布
Phase 11	Runtime Cognition (Cognition/Fabric/Compiler 2.0/AI Coprocessor)	✅ 已发布
Phase 12	Autonomous Scientific Civilization Runtime (Evolution/Ecology/Swarm/Scientist)	✅ 已发布
Phase 13	Scientific Universe Runtime (Physics/Time/Thermo/Life/Observers/Domain Earthquake)	✅ 已发布
Phase 14	AI Observer API — 只读 snapshot 接口层 (31 快照类型, 6 crates)	✅ 已发布

---

端到端数据流

parse_formula("x = (40 * AU * (1 - kappa)) / c")
  → lex → Token[]
  → parse → Expr::Assign
  → to_ir() → Vec<IRNode>
  → graph.insert_ir() → propagate_dirty() → get_value("x") = 42.0

---

测试状态

范围	测试数	状态
Runtime (Phase 9)	209	✅
HPC Solver (Phase 5)	19	✅
Advanced GPU (Phase 4)	6	✅
GPU Newton (Phase 4)	5	✅
GPU Batch (Phase 4)	5	✅
Runtime Evolution (Phase 12.1)	32	✅
Scientific Knowledge (Phase 12.2)	24	✅
Compiler Ecology (Phase 12.3)	20	✅
Fabric Swarm (Phase 12.4)	20	✅
Runtime Scientist (Phase 12.5)	22	✅
Runtime Physics (Phase 13)	102	✅
Runtime Time (Phase 13)	65	✅
Runtime Thermodynamics (Phase 13)	43	✅
Runtime Life (Phase 13)	49	✅
Runtime Observers (Phase 13)	63	✅
Domain Earthquake (Phase 13)	68	✅
编译 warning	—	⚠️ 少量(未清理)

---

快速开始

# 编译 Runtime
cd ~/sym-nebula

# 运行全部 Runtime 测试
cargo test -p symnebula-runtime

# 编译 Python Bridge
cargo check -p symnebula-runtime --features python

# 编译整个工作区
cargo check --workspace

---

项目结构

sym-nebula/
├── runtime/                 # 主 Runtime crate (Phase 6-9)
│   ├── src/
│   │   ├── parser/          # Formula lexer + parser + AST + IR
│   │   ├── graph/           # IRNode + ConstraintGraph + RPN fusion
│   │   ├── stream/          # Event bus (tokio broadcast)
│   │   ├── serialization/   # Snapshot save/load (serde JSON)
│   │   ├── session/         # RuntimeSession + API + External API
│   │   ├── ffi/             # C ABI layer (extern "C")
│   │   ├── plugins/         # SolverPlugin + NodePlugin + ModuleLoader
│   │   ├── sandbox/         # Capability sandbox
│   │   ├── registry/        # RuntimeRegistry (聚合)
│   │   ├── sparse/          # CSR/COO sparse matrix + Jacobian
│   │   ├── optimizer/       # DAG optimizer (folding/fusion/partition)
│   │   ├── tensor/          # N-D Tensor + autodiff + solver
│   │   ├── pde/             # PDE solvers (Heat/Wave/Poisson/NS/FEM)
│   │   ├── compiler/        # Lowering → GPU/SIMD Codegen → JIT
│   │   ├── dsl/             # Tensor/PDE DSL grammar + parser
│   │   ├── gpu/             # GPU kernel fusion (WGSL)
│   │   ├── observatory/     # Runtime visualization (9 viewers)
│   │   ├── scheduler/       # Adaptive GPU/CPU/Remote scheduler
│   │   ├── cluster.rs       # Cluster node management
│   │   ├── fabric.rs        # Network fabric
│   │   ├── distributed_scheduler.rs
│   │   ├── persistent_world.rs
│   │   ├── network/         # TCP/UDP transport
│   │   ├── storage/         # Persistent storage
│   │   ├── monitor/         # System monitor
│   │   ├── wasm/            # WASM stub
│   │   ├── python_bridge.rs # pyo3 Python bindings
│   │   └── lib.rs
│   └── Cargo.toml
├── symmath/                # 历史 GPU/分布式求解器 (Phase 4-5)
│   └── crates/
│       ├── gpu_batch_solver/
│       ├── gpu_newton_solver/
│       ├── gpu_newton_solver_advanced/
│       ├── distributed_cluster_solver/
│       └── distributed_cluster_solver_hpc/
├── runtime_evolution/      # Phase 12.1 — Runtime Evolution Engine
├── scientific_knowledge/   # Phase 12.2 — Scientific Knowledge Graph
├── compiler_ecology/       # Phase 12.3 — Autonomous Compiler Ecology
├── fabric_swarm/           # Phase 12.4 — Autonomous Fabric Swarm
├── runtime_scientist/      # Phase 12.5 — Scientific AI Runtime Scientist
├── runtime_physics/        # Phase 13.1 — CausalGraph + Seismic Runtime (+ api.rs)
├── runtime_time/           # Phase 13.2 — Timeline + CausalClock + Rollback (+ api.rs)
├── runtime_thermodynamics/ # Phase 13.3 — EnergyBudget + Equilibrium (+ api.rs)
├── runtime_life/           # Phase 13.4 — Organism + Ecosystem + Extinction (+ api.rs)
├── runtime_observers/      # Phase 13.5 — Observer + Telescope + Dashboard (+ api.rs)
├── domain_earthquake/      # Phase 13.7 — Seismic domain model + ETAS + Guangxi (+ api.rs)
├── core/                   # 历史核心引擎 (legacy)
├── macros/                 # proc-macro crate
├── docs/
│   ├── PHASE6_ROADMAP.md
│   ├── PHASE7_ROADMAP.md
│   └── PHASE8_ROADMAP.md
└── README.md

---

设计原则

1. Runtime First, AI Optional — AI/LLM 是外部客户端，不侵入核心
2. Runtime deterministic — 相同公式 + 输入 → 相同输出
3. Plugin ≠ Core — 求解器、节点类型、模块都是插件
4. Sandbox Everywhere — 每个插件绑定 RuntimeCapability
5. Context-window friendly — export_context_summary() 提供摘要而非完整 DAG
6. AI = Observer, Not Mutator — AI 只能通过只读 snapshot API 观察宇宙状态（各 crate 的 src/api.rs），不能直接修改内部状态。所有写操作须经 RuntimeSession 事务隔离

---

开源协议 | License

GNU General Public License v3.0

Copyright (c) 2026 NebulaMind