xstate-statemachine 0.4.2

A robust Python library for parsing and running XState JSON state machines.

🚦 XState - StateMachine for Python

The Definitive ← Seriously → Guide to Bullet-Proof State Machines

“A good state machine is like a map: once you have it, you’ll never get lost again.”

---

📜 Table of Contents

1. Introduction
2. Why State Machines?
3. Event‑Driven Architecture 101
4. What is XState-StateMachine?
5. Key Features
6. Installation
7. Quick Start
8. The State Machine Philosophy
9. Visual-First Development
10. Anatomy of an XState JSON Blueprint
11. States — Atomic, Compound, Parallel, Final
12. Transitions & Events
13. Actions, Guards & Services
14. Context - The Machine’s Memory
15. Declarative Timers (after)
16. The Actor Model
17. Architectural Patterns
18. Synchronous vs Asynchronous Execution
19. Debugging & Visualization
20. CLI: Boilerplate Generation - Aliases & Short Flags - Hierarchical Machine Generation - One vs Two Files - Sync vs Async Templates
21. Upgrade Notes: 0.4.0 → 0.4.1
22. API Reference
23. Advanced Concepts
24. Best Practices
25. FAQ
26. Roadmap
27. Contributing
28. License

---

🏁 Introduction

Welcome to XState-StateMachine for Python, a robust, async-ready, and feature-complete library for building, parsing, and executing state machines and statecharts defined in XState-compatible JSON.

Whether you’re a junior dev struggling with spaghetti if/else trees 🌱 or a senior architect with half a century of scars and stories 🦖, this README is crafted to be your Bible. By the time you finish, you’ll know why state machines matter, how to model them visually, and exactly what code to write to run them in production-grade Python.

---

❓ Why State Machines?

1. Eliminate Impossible States 🧹 Four boolean flags ➡ 16 possible combinations. How many are valid? State machines guarantee you’ll never enter a “loading + error” paradox again.

2. Explode Complexity—In a Good Way 💥 Complexity will happen. Put it in a graph where it’s explicit, testable, and visualized—rather than hidden in nested conditionals.

3. Single Source of Truth 🔑 Your JSON blueprint declares every state, event, and transition. No surprises lurking in random helper functions.

4. Self-Documenting 📚 A statechart is the documentation. No more stale flow-charts stuck in a Confluence graveyard.

5. Safer Concurrency 🛡️ Parallel states and the Actor Model let you reason about multi-threaded logic without race-condition nightmares.

---

⚡ Event‑Driven Architecture 101

“Everything is an event.” — That’s the mindset shift.

Most real systems are reactive: something happens → you react. Event‑Driven Architecture (EDA) makes that explicit and first‑class.

🧠 Core Ideas

• Emit facts, don’t call functions. Instead of “do X then Y”, you announce that “X happened”, and anyone interested can react.
• Decoupling by default. Producers don’t know who (or how many) consumers will handle an event.
• Time becomes visible. Events arrive over time; your logic becomes a timeline, not a call stack.

🧩 How XState-StateMachine fits EDA

EDA Need	Library Feature	Why it helps
“Everything is an event”	on, after, done.invoke.*, error.platform.*, synthetic entry./exit.	One uniform pipeline; nothing is “special-cased”.
Deterministic reactions	Guards & Actions	Pure decisions (guard) vs. side effects (action).
Temporal logic	Declarative after timers	Timeouts, retries, heartbeats—no manual sleep()s.
Isolation & composition	Actor model (spawn_*)	Each actor is its own event-loop bubble.
Observability	Plugins & snapshots	Tap and record every heartbeat for logs/tests.

🗺️ Tiny Example: “Upload finished” flow

sequenceDiagram
    participant UI
    participant SM as State Machine
    participant API

    UI->>SM: START_UPLOAD
    SM->>API: invoke uploadFile()
    API-->>SM: done.invoke.uploadFile (data)
    SM->>UI: SHOW_SUCCESS

No direct await upload() in UI logic; the machine emits/receives events and drives the UI.

📦 Example Payload Flow

USER_CLICKED_SUBMIT
           │
           ▼
   [guard: isValid]  ────✗→  (event dropped)
           │
           ▼
   actions: saveDraft → invoke apiCall
                          │
                          ├── done.invoke.apiCall  → actions: cache, target: success
                          └── error.platform.apiCall → actions: setError, target: failure

✅ Why developers love it

• Clear “When X happens, go to Y” rules—no hidden callbacks.
• Visual diagrams match exactly what runs.
• Tests can fire events in order and assert outcomes—no flaky timing.

Takeaway: State machines are a great way to implement EDA. Events, timers, and actors are the primitives; this library wires them up so you don’t have to.

---

✨ What is XState-StateMachine?

• A Pythonic runtime for the world-famous XState architecture.
• 100 % JSON-spec-compatible, so you can design your chart in the Stately editor and run it untouched.
• Async first (asyncio), yet ships a fully-featured, blocking SyncInterpreter for CLI tools or tests, complete with support for after timers and a threaded actor model.
• Packed with goodies: hierarchy, parallelism, invoke, after, timers, actors, auto-binding logic loaders, plugin hooks, diagram generators, and more.

TL;DR — If you know XState in JS, everything 👉 “just works” in Python. If you don’t, keep reading—this guide is for you.

---

🚀 Key Features

Feature	Why You Care	New in 0.4.1?	Best For
100% XState JSON Compatibility	Design visually, export JSON, run in Python.		Teams that collaborate in the Stately Editor.
Async & Sync Interpreters	One config, two runtimes (asyncio or blocking).	⬆️ SyncInterpreter now non‑blocking for actor spawn & multi‑timers	Apps that need to run in CLI scripts and async servers.
Multiple after Timers per State (Sync too!)	Fine-grained timing logic without hacks.	✅	Complex timeout/backoff flows in synchronous environments.
True Actor Model (spawn_ actions)*	Compose systems from smaller machines; isolate concurrency.	⬆️ Sync spawn now uses background threads	IoT fleets, game entities, micro‑workflows.
Hierarchical & Parallel States	Model nested flows or concurrent regions explicitly.		UIs, wizards, orchestration logic.
Declarative invoke Services	Async tasks with automatic cancellation & onDone/onError.		API calls, DB work, long-running jobs.
Smart CLI Code Generator (xsm)	Zero-boilerplate logic/runner stubs from JSON.	✅ Aliases, hierarchy detection, safer prompts, smarter paths	Fast scaffolding, consistent team patterns.
Auto Logic Binding (LogicLoader)	Drop in modules/classes; names are auto-wired.	⬆️ No special-casing spawn_*	Big projects with many logic files.
Deep Plugin Hooks	Observe guards, services, transitions in detail.		Telemetry, tracing, testing.
Snapshots & Restoration	Time-travel debugging, crash recovery.		Long-lived workflows, CI golden tests.
Diagram Exporters	Mermaid / PlantUML from code.		Up-to-date docs in CI/CD.

🔥 Headline for 0.4.1 SyncInterpreter grows up: non‑blocking actor spawning in threads, multiple concurrent after timers per state, and a cleaner shutdown sequence. CLI super-powers: xsm, subcommand aliases, interactive parent detection, stricter arg parsing, and more robust generated runners.

---

🛠️ Installation

# 1️⃣ Create & activate a virtual env  (recommended)
python -m venv venv
source venv/bin/activate           # Windows → venv\Scripts\activate

# 2️⃣ Install the library
pip install xstate-statemachine

Requirements: Python 3.8 +

---

⚡ Quick Start

A lightspeed tour: toggle a light 💡—the “Hello World” of state machines.

1. Blueprint (light_switch.json)

{
  "id": "lightSwitch",
  "initial": "off",
  "context": {
    "flips": 0
  },
  "states": {
    "off": {
      "on": {
        "TOGGLE": {
          "target": "on",
          "actions": "increment_flips"
        }
      }
    },
    "on": {
      "on": {
        "TOGGLE": {
          "target": "off",
          "actions": "increment_flips"
        }
      }
    }
  }
}

2. Logic (light_switch_logic.py)

import logging
from typing import Dict
from xstate_statemachine import Interpreter, Event, ActionDefinition

def increment_flips(i: Interpreter, ctx: Dict, e: Event, a: ActionDefinition):
    ctx["flips"] += 1
    logging.info(f"🔀 Switch flipped {ctx['flips']} time(s).")

3. Runner (main.py)

import asyncio
import json
import light_switch_logic # Use a standard import
from xstate_statemachine import create_machine, Interpreter

async def main():
    with open("light_switch.json") as f:
        config = json.load(f)

    # Pass the imported module object directly for auto-discovery. This is cleaner.
    machine = create_machine(config, logic_modules=[light_switch_logic])

    interpreter = await Interpreter(machine).start()
    await interpreter.send("TOGGLE")
    await interpreter.send("TOGGLE")
    await interpreter.stop()

asyncio.run(main())

Output

INFO 🔀 Switch flipped 1 time(s).
INFO 🔀 Switch flipped 2 time(s).

Boom—no if current_state == "on" anywhere. 🎉

A Very Simple Sync Example: The Toggle Switch

Here is the most basic example of a synchronous state machine. It has only two states (on and off) and one event (TOGGLE). It perfectly illustrates how the SyncInterpreter processes events immediately.

1. The Blueprint: toggle_switch.json

This JSON defines the structure. It starts off, and the TOGGLE event switches it to on (and vice-versa), running an action called increment_toggles each time.

{
  "id": "toggleSwitch",
  "initial": "off",
  "context": {
    "toggleCount": 0
  },
  "states": {
    "off": {
      "on": {
        "TOGGLE": {
          "target": "on",
          "actions": [
            "increment_toggles"
          ]
        }
      }
    },
    "on": {
      "on": {
        "TOGGLE": {
          "target": "off",
          "actions": [
            "increment_toggles"
          ]
        }
      }
    }
  }
}

2. The Logic and Runner: main_sync.py

For maximum simplicity, we'll define the logic and the simulation in the same file.

import json
import logging
from typing import Dict, Any

from xstate_statemachine import (
    create_machine,
    SyncInterpreter,
    MachineLogic,
    Event,
    ActionDefinition
)

# --- Basic Setup ---
logging.basicConfig(level=logging.INFO, format="[%(levelname)s] %(message)s")

# --- 1. The Logic (Action) ---
# This is the single action our machine will execute.
def increment_toggles(i: SyncInterpreter, ctx: Dict, e: Event, a: ActionDefinition) -> None:
    """Action to increment the toggle count in the context."""
    ctx["toggleCount"] += 1
    logging.info(f"💡 Light is now {i.current_state_ids}. Toggle count: {ctx['toggleCount']}")

# --- 2. The Simulation ---
def run_simple_toggle():
    """Initializes and runs the toggle switch simulation."""
    print("\n--- 💡 Simple Synchronous Toggle Switch ---")

    # Load the machine configuration from the JSON file
    with open("toggle_switch.json", "r") as f:
        config = json.load(f)

    # Explicitly bind the action name "increment_toggles" from the JSON
    # to our Python function.
    logic = MachineLogic(
        actions={"increment_toggles": increment_toggles}
    )

    # Create the machine and the synchronous interpreter
    machine = create_machine(config, logic=logic)
    interpreter = SyncInterpreter(machine)

    # Start the machine. It enters the 'off' state.
    interpreter.start()
    logging.info(f"Initial state: {interpreter.current_state_ids}")

    # Send the first event. The machine transitions to 'on'
    # and the 'increment_toggles' action runs before this line finishes.
    print("\n--- Toggling ON ---")
    interpreter.send("TOGGLE")

    # Send the second event. The machine transitions back to 'off'
    # and the action runs again.
    print("\n--- Toggling OFF ---")
    interpreter.send("TOGGLE")

    # Stop the machine
    interpreter.stop()
    print("\n--- ✅ Simulation Complete ---")

if __name__ == "__main__":
    run_simple_toggle()

Expected Output

When you run main_sync.py, you will see the following output, demonstrating that each send call completes its work before the next print statement is executed:

--- 💡 Simple Synchronous Toggle Switch ---
[INFO] Initial state: {'toggleSwitch.off'}

--- Toggling ON ---
[INFO] 💡 Light is now {'toggleSwitch.on'}. Toggle count: 1

--- Toggling OFF ---
[INFO] 💡 Light is now {'toggleSwitch.off'}. Toggle count: 2

--- ✅ Simulation Complete ---

---

🧭 The State Machine Philosophy

Definition ↔ Implementation separation is the super-power.

• Definition (.json) 😇 — declares what can happen.

  • Finite states
  • Events
  • Valid transitions
  • Timers, services, hierarchy

• Implementation (.py) 🛠️ — implements how it happens.

  • Fetch an API
  • Write a file
  • Update UI

Because the graph never mutates, every team-mate sees the same reality. Changing business rules is as easy as editing JSON and re-running tests—logic stays untouched.

---

🎨 Visual-First Development

1. Design in the Stately Editor → drag states, draw arrows.
2. Export to JSON (one click).
3. Run with create_machine(config) in Python.
4. Simulate inside Stately or via Python tests—they behave identically.

Why It Rocks:

• Stakeholder Friendly — Product managers & QA can read and play with the diagram.
• Zero Drift — Diagram is the code. Update one, you update both.
• Faster On-Boarding — New hires grok the flow in minutes, not days.

---

🖼️ Designing with Stately Visual Editor

“If a picture is worth a thousand words, a statechart is worth a thousand unit-tests.”

The Stately Visual Editor → is the single most productive tool in the XState ecosystem. It lets you draw your machine, simulate it in real‑time, and export a perfectly–valid JSON blueprint that runs unchanged in xstate‑statemachine for Python.

🔑 Why the Editor Matters

Benefit	What it Means for You
WYSIWYG Modelling	Drag‑and‑drop states, draw transitions, and tweak guards—no JSON eye‑strain.
Instant Simulation	Play events, watch timers fire, and inspect context mutations live before writing Python code.
Team Collaboration	Share a link; PMs and QA can see the flow and leave comments, killing a whole back‑and‑forth thread of screenshots.
Source of Truth	The exported JSON is the code—zero drift between docs and implementation.
Version Control Friendly	Download the JSON (or embed it in your repo) and diff it like any other text asset.

🛠️ Typical Workflow

1. Sketch the high‑level flow on a whiteboard (or directly in the editor).

2. Model it in the Stately Editor:

   • Add states (double‑click canvas)
   • Create transitions (drag arrow from one state to another)
   • Configure events, guards, actions, after timers, and invoke services in the side‑panel UI.

3. Simulate: hit ▶️, dispatch events, and watch the visual debugger update in real time.

4. Export → “Machine JSON” (⚙️ menu → Export → Machine JSON). Save as my_machine.json in your project’s statecharts/ folder.

5. Run it with:

   from xstate_statemachine import create_machine, Interpreter
   import json, asyncio
   
   cfg = json.load(open("statecharts/my_machine.json"))
   machine = create_machine(cfg, logic_modules=[my_logic])
   await Interpreter(machine).start()
   

6. Iterate: tweak the diagram, re‑export, rerun tests. Your Python logic remains untouched.

🎨 Pro Tips for Power Users

Tip	Shortcut / Action
Multi‑Select	Shift + Click or drag marquee to move/align groups of states.
Quick Transition	Hold A, click source state, then click target state.
Relative Targets	Double‑click a self‑transition arrow to toggle between internal and external (re‑entry) semantics.
Context Visualisation	In the Simulate tab, expand Context to live‑edit values while the machine is running—great for guard testing.
Export Diagram	Export → SVG / PNG to embed in GitHub docs; keep diagrams and JSON in‑sync ✨.
Embed Gist	Publish the machine as a sharable, live gist you can link in PR descriptions.

---

🧬 Anatomy of an XState JSON Blueprint

Every machine is a tree of StateNodes. Let’s break down the top-level keys:

Key	Type	Description
id	string	Unique machine ID, root of every absolute state path.
initial	string	State where the interpreter starts.
context	object	Mutable “memory” available to every action/guard.
states	object	Map of state → StateNode definition.
on	object	Global event handlers (catch-all).

Example from your files: The flightBooking.json machine has a root id of "flightBooking".

Example from your files: The ciCdPipeline.json defines an initial context to track the state of a deployment process:

"context": {
  "build_artifact": null,
  "commit_hash": null,
  "deployment_url": null,
  "error": null,
  "scan_results": null,
  "test_results": null
}

Target Resolution: How the Machine Finds the Next State

When you define a transition with a target, the library uses a powerful resolution mechanism to find the destination state. This allows for flexible and intuitive state navigation.

• Sibling State (Most Common): If you provide a simple name, the interpreter looks for a state with that name within the same parent.

  "green": {
    "on": { "TIMER": "yellow" } // Looks for "yellow" alongside "green"
  },
  "yellow": {},
  "red": {}
  

• Child State (Dot Notation): To target a descendant state, use dot notation.

  "on": {
    "GO_TO_DEEP_STATE": "parent.child.grandchild"
  }
  

• Relative Path (Leading Dot): A leading dot (.) makes the path relative to the parent of the current state. This is extremely useful for sibling-to-sibling transitions inside a compound state.

  "parent": {
    "initial": "child1",
    "states": {
      "child1": { "on": { "NEXT": ".child2" } }, // Correctly targets parent.child2
      "child2": {}
    }
  }
  

• Absolute Path (Leading Hash): A leading hash (#) makes the path absolute from the root of the machine, using the machine's id. This is the safest way to target a state from a deeply nested location without ambiguity.

  "deeply": {
    "nested": {
      "state": {
        "on": {
          "GO_HOME": "#myMachine.idle" // Always goes to the top-level idle state
        }
      }
    }
  }
  

🛣️ Edge‑Cases & Safety Nets

Target Syntax	Resolved To	When to Use
"."	Parent state of the current state  ▸ if already at the root, it resolves to the root itself.	Jump back one level without hard‑coding the parent’s ID. Handy inside deeply nested compound states.
a..b  or  state.	❌ Invalid — the library rejects any target that contains empty path segments (double dots .. or a trailing dot .) and raises StateNotFoundError.	Typos happen! The explicit error prevents silent mis‑navigation and keeps your diagrams truthful.

💡 Tip: When debugging a mysterious StateNotFoundError, check for an accidental double‑dot or dangling dot in your target strings.

---

🏛️ States — Atomic • Compound • Parallel • Final

1️⃣ Atomic

No children. Think “leaf node.”

"idle": {
  "on": { "FETCH": "loading" }
}

2️⃣ Compound

Has its own initial + states.

Example from your files: In installWizard.json, the "configuring" state is a compound state that contains its own sub-machine for handling the configuration steps.

"configuring": {
  "initial": "network",
  "onDone": "installing",
  "states": {
    "network": {
      "on": {
        "SUBMIT_NETWORK": "database"
      }
    },
    "database": {
      "on": {
        "SUBMIT_DATABASE": "admin_user"
      }
    },
    "admin_user": {
      "on": {
        "SUBMIT_ADMIN": "config_complete"
      }
    },
    "config_complete": {
      "type": "final"
    }
  }
}

3️⃣ Parallel

All child regions active at once. Perfect for independent subsystems.

Example from your files: The smartHome.json machine uses a parallel state at its root to manage lighting, climate, and security as independent, concurrent regions.

"dashboard": {
  "type": "parallel",
  "states": {
    "notifications": {
      /* handles toasts */
    },
    "socket": {
      /* websocket connect/close */
    },
    "theme": {
      /* dark / light */
    }
  }
}

4️⃣ Final

Signals completion to parent; triggers onDone transition.

"success": { "type": "final" }

---

🔄 Transitions & Events

🌊 Event Lifecycle & Synthetic Events

A state machine lives and breathes events. Besides the ones you dispatch, the runtime forges its own messages and even loops internal “always” transitions until the graph stabilises. Understanding these moving parts lets you write bullet‑proof tests, guards and plugins. 🔍

---

1️⃣ .send() – One API, Three Input Flavours

What you call	What the helper returns	Notes
service.send("CLICK", x=1)	Event(type="CLICK", payload={"x": 1})	Snack‑size syntax
service.send({"type": "CLICK", "x": 1})	ditto	Handy when forwarding raw JSON
service.send(Event("CLICK", {"x": 1}))	unchanged	Already a proper Event

BaseInterpreter._prepare_event does the coercion, so every path into the interpreter is consistent & type‑safe. ✔️

---

2️⃣ Runtime‑Generated (Synthetic) Events

Pattern	✨ When it fires	Typical purpose
entry.<stateId>	Right before a state’s entry actions run	Side‑effect hooks, analytics
exit.<stateId>	After exit actions finish	Cleanup metrics, audit
___xstate_statemachine_init___	Once, at machine start‑up	Kick‑start transient guards
after.<delay>.<stateId>	after { "<delay>": … } timer expires	Declarative timeouts / polling
done.state.<stateId>	A compound / parallel state reaches all its finals	Bubble completion upward
done.invoke.<src>	An invoked service returns successfully	Happy‑path transitions
error.platform.<src>	Invoked service raised / rejected	Failure branch

Because they are regular events you can:

await interp.send("after.5000.flightBooking.loading")   # force timeout in tests
plugin.on_event_received = lambda _, e: print(e.type)

Quick Example: onDone from a Parallel Parent

"checkout": {
  "type": "parallel",
  "states": {
    "shipping": { "type": "final" },
    "billing":  { "type": "final" }
  },
  "onDone": "receipt"  // fires when BOTH regions hit their finals
}

---

3️⃣ Transient (“Always”) Transitions ""

An empty‑string event ("") models automatic logic that should run immediately after a state becomes active:

"checking": {
  "on": {
    "": [
      {
        "guard": "isValid",
        "target": "approved"
      },
      {
        "target": "rejected"
      }
    ]
  }
}

• Both interpreters keep looping while optimal_transition.event == "": … (Interpreter._run_event_loop, SyncInterpreter._process_transient_transitions) until no guard passes.
• Guards must be pure & synchronous — they run potentially many times per event cycle.
• Great for conditional redirects, validation gates and hierarchical “initial” logic.

---

📝 Cheat‑sheet

You           Engine                          What you observe
──────────── ─────────────────────────────── ─────────────────────────────────
service.send("CLICK")      ─────▶  CLICK
state entry                ─────▶  entry.someState
timer 2 s later            ─────▶  after.2000.someState
service success            ─────▶  done.invoke.fetchData
compound finished          ─────▶  done.state.parent.compound

Clarification — arrows show the event type, not the payload In the cheat‑sheet diagram, service.send("CLICK", x=1) ─────▶ CLICK means the runtime receives an event whose type is "CLICK". Any payload you pass (e.g. x=1) still travels inside the event object (Event.type == "CLICK", Event.payload == {"x": 1}); it’s just omitted for brevity.

Now you can assert, spy and debug every heartbeat of your machine. 🎉

• Event-driven — under on.
• Time-driven — under after.
• Done/Error — from invoke services → auto-events done.invoke.<src> & error.platform.<src>.

"loading": {
  "invoke": {
    "src": "fetchData",
    "onDone": {
      "actions": "cacheData",
      "target": "success"
    },
    "onError": {
      "actions": "setError",
      "target": "failure"
    }
  },
  "after": {
    "5000": {
      "target": "timeout"
    }
    // If it hangs for 5 s
  }
}

📨 How service results travel back into the machine

When an invoked service (sync or async) finishes successfully, its return value is baked into a DoneEvent:

type =  "done.invoke.<serviceId>"
data =  <return value of your Python function>

That payload is now available to guards & actions via event.data — use it to make decisions or stash the result in context.

# guards.py
def is_valid_response(ctx, event):
    # event.data is whatever the service returned 🙌
    return event.data.get("status") == 200

# actions.py
def store_payload(i, ctx, event, a):
    ctx["payload"] = event.data["body"]

"loading": {
  "invoke": {
    "src": "fetchData",
    "onDone": [
      { "target": "success", "guard": "is_valid_response", "actions": "store_payload" },
      { "target": "failure" }
    ],
    "onError": "failure"
  }
}

🔎 Tip: In unit tests you can stub the service to return a canned object and assert that ctx["payload"] matches it, without hitting the network. Fast & deterministic! 🧪

Internal vs External Transitions & The reenter Flag

When an event matches a transition, the library distinguishes between two types: internal and external.

• Internal Transition: An internal transition only executes its actions and does not exit or re-enter the current state. This is useful for updating context without resetting the state's timers or invoked services. This occurs when a transition has no target.

• External Transition: An external transition will exit the current state (running exit actions, cancelling timers/services) and re-enter a new state (or the same one), running all entry actions. This occurs whenever a transition has a target.

The reenter Flag for Self-Transitions (New in 0.4.2)

By default, a self-transition (where the target is the same as the source state) is internal. However, you can force it to be external by adding "reenter": true. This provides explicit control, aligning with XState v5.

Transition Config	Behavior	Use Case
{"actions": "update_ctx"}	Internal (no target)	Update context without resetting the state.
{"target": "myState"}	Internal (self-target, reenter is false by default)	Same as above. The machine remains in myState without exit/entry actions.
{"target": "myState", "reenter": true}	External (self-target, explicit reenter)	Force a full exit/re-entry of myState to reset its timers or re-run entry actions.

Example:

Consider a state that needs to reset an inactivity timer whenever the user performs an action.

"active": {
  "initial": "idle",
  "states": {
    "idle": {
      "after": {
        "5000": { "target": "#myMachine.timedOut" }
      },
      "on": {
        "USER_ACTIVITY": {
          "target": "idle",
          "reenter": true // This forces an exit/re-entry, resetting the 5s timer
        }
      }
    }
  }
}

---

🛠️ Actions, Guards & Services

✅ Good – deterministic & side‑effect‑free

def can_retry(ctx, event) -> bool:
    return ctx["attempts"] < 3

🚫 Bad – asynchronous

async def remote_rule(ctx, event):
    result = await fetch_flag()            # blocking the event loop = 💥
    return result == "ALLOW"

🚫 Bad – non‑boolean return

def non_bool(ctx, event):
    return "yes" if ctx["foo"] else ""     # Truthy string! ⚠️

While "yes" passes today, explicit True/False is required for readability and future compatibility.

---

Tip 💡 – Keep Them 100 % Pure

• No logging inside guards – use an action instead.
• No mutation of ctx. Guards run many times (transients!), so mutating state here creates elusive bugs.

---

Now your transitions obey the Law of Least Surprise: one question, one crisp answer, synchronously. 🏁

Kind	When Runs	Signature	Return
Action	On entry/exit/transition	(interp, ctx, event, action_def)	None
Guard	Before transition decision	(ctx, event)	bool
Service	Inside invoke (async or sync)	(interp, ctx, event)	value or raise

🔥 Error Handling (Actions / Guards / Services)

TL;DR

Component	If it raises…	What the interpreter does	How to observe / recover
Guard	Any Exception	Treated as False (transition candidate fails), error is logged; the interpreter keeps evaluating the rest of the transitions. If none pass, the event is dropped.	Use LoggingInspector or a custom plugin (on_guard_evaluated) to surface failures.
Action	Any Exception	Logs the error, skips the remaining actions in that list, continues processing. In SyncInterpreter the exception is re‑raised to the caller of .send(). In Interpreter it’s captured inside the event-loop task (logged, does not crash the loop).	Wrap .send() in try/except for sync; add a plugin or await send() and inspect logs for async.
Service (invoke)	Exception / rejection	Fires the configured onError transition with event.data = error; if no handler exists, the error bubbles to the parent (or is logged).	Handle onError branches explicitly; inspect event.data for details.

# Guard example: an exception becomes "False", not a crash
def might_blow_up(ctx, event):
    return 1 / ctx["divisor"] > 0   # ZeroDivisionError → guard == False

# Action example: catching sync errors
try:
    sync_service.send("SAVE")
except Exception as exc:
    logger.error("SAVE failed: %s", exc)

🐍 Snake → Camel Autowiring 🐫 The loader automatically converts snake_case Python function names to camelCase keys expected in your JSON. No manual mapping needed — simply define:

def increment_flips(i, ctx, e, a): ...

…and reference it in JSON as:

"actions": "incrementFlips"

The helper logic_loader._snake_to_camel() does the heavy lifting, so you stay idiomatic in Python and compliant with XState’s camel‑cased world. ✨

ℹ️ Automatic Logic Discovery binds JSON names to Python callables by convention (snake_case ⇌ camelCase). Anything unmatched raises ImplementationMissingError.

---

🧠 Context - The Machine’s Memory

A plain dict shared across all states. Mutate it inside actions and services; read-only in guards.

"context": {
  "retryCount": 0,
  "payload": null,
  "error": null
}

def increment_retry(i, ctx, e, a):
    ctx["retryCount"] += 1

Rule of thumb 🧘: Pure guards & deterministic actions → easier tests.

---

⏰ Declarative Timers (after)

Key idea: you declare intent, not implementation. The interpreter owns the stopwatch so you don’t have to.

What you write	What the engine does
after: { "500": "blink" }	Starts a 500 ms countdown every time the state is entered
after: { "1000": { "target": "retry", "actions": "backoff" } }	Schedules an internal event, then cancels it automatically if the state exits early
after: { "…": { …, "guard": "stillRelevant" } }	Evaluates guard right before firing—handy for stale timers

SyncInterpreter parity (0.4.1): Everything you see here—including multiple after entries per state—now works in the synchronous engine. Each after gets a dedicated thread-backed timer and a unique ID for cancellation. No extra code from you.

SyncInterpreter parity addendum – guard timing: In SyncInterpreter, exactly like in the async Interpreter, any guard on an after transition is evaluated when the timer fires, not when it’s scheduled on state entry. If the guard returns False at fire time, that particular delayed transition is skipped (other after entries can still fire), and no state change occurs.

"retrying": {
  "after": {
    "1000": { "target": "fetching", "guard": "fewAttempts" },
    "5000": { "target": "abandon" }
  }
}

def fewAttempts(ctx, event):
    # evaluated only when the 1000ms timer actually fires
    return ctx["attempts"] < 3

Anatomy of an after Event

after(<delay>)#<absoluteStateId>
└──── ─────── ── ───────────────
   |     |          |
   |     |          ↳ state that was active when timer started
   |     |
   |     ↳ milliseconds (integer)
   |
   ↳ literal string “after”

Knowing the auto‑generated type helps if you need to assert or spy on timers in tests:

await interp.send("START_LOAD")
await asyncio.sleep(0.01)                  # Allow event loop to process
assert "loading" in interp.current_state_ids

# Fast‑forward virtual time by monkey‑patching loop.time() — or simpler:
await asyncio.sleep(5.1)                   # let after(5000) fire
assert "timeout" in interp.current_state_ids

⚡ Multiple Timers in One State

Attach several after clauses—think progressive back‑off:

"retrying": {
  "entry": "incAttempts",
  "after": {
    "1000": {
      "target": "fetching",
      "guard": "fewAttempts"
    },
    "5000": {
      "target": "giveUp",
      "actions": "alertOps"
    }
  }
}

If fewAttempts returns false, the 1‑second timer is ignored; the 5‑second timer is still pending.

⏱️ Looping Timers (Self‑Transition)

"flashing": {
  "after": {
    "250": {
      "actions": "toggleLED",
      "target": ".flashing"
    }
  }
}

Because the target is relative (.flashing), the state re‑enters itself, creating a persistent 250 ms blink. The interpreter guarantees only one live timer at a time—previous handles are cancelled on re‑entry.

Cancellation & Clean‑Up

Leaving a state always cancels its timers.

• No memory leaks
• No stray events after the user navigates away
• Predictable “at‑most‑once” semantics

Testing Timers 🧪

1. Unit – patch the event loop clock (pytest‑asyncio’s advance_time) and assert the synthetic event.
2. Integration – run under a virtual loop with controlled time.
3. Snapshots – diff context before & after the timer fires.

Common Pitfalls & Remedies

Pitfall	Symptom	Remedy
Timer never fires	Guard returns False; or state changed before delay	Log guard or use LoggingInspector
Fires twice	Re‑entering state via different absolute ID in hierarchy	Use absolute target ("#machine.state") or guard
Delay starts late	Long CPU loop in entry action	Make actions async & yield await asyncio.sleep(0)

🗂️ TaskManager – Zero‑Leak Guarantees

Every after timer ⏱️ and each invoke service 📞 is tracked by the TaskManager: the async Interpreter wraps them in asyncio.Task, while the SyncInterpreter uses threading.Timer/threading.Thread; both are registered per‑state and cancelled on exit.

stateId ─┬─ after‑5000 timer        ─┐
         ├─ after‑10000 timer  ──────┤───▶ TaskManager.add(owner_id, task)
         └─ invoke.fetchData task ───┘

Why it matters:

⭐ Benefit	How it works
No orphaned coroutines	When a state exits, the interpreter calls TaskManager.cancel_by_owner(state.id), which iterates over every recorded task, task.cancel()s them, and awaits graceful shutdown.
Memory‑safe	The internal map is cleaned after cancellation, so tasks don’t linger in RAM.
Race‑condition free	Timers or services started in a state cannot out‑live that state; you’ll never receive a late “after” event for something that’s no longer on screen.

🔒 Guarantee: If your JSON says “when I leave loading, kill the fetch”, the engine obeys—you write zero cancellation code.

---

🧪 White‑Box Testing Helper

Need to assert that a timer or service is (or isn’t) running?

tasks = interpreter.task_manager.get_tasks_by_owner("search.loading")
assert len(tasks) == 1          # the after(8000) timeout

get_tasks_by_owner(owner_id) returns a copy of the task set, so your test can inspect without risking accidental mutation.

---

⌛ Bottom line: Declarative timers and invokes stay tidy, deterministic and resource‑safe—no leaks, no zombies, no surprises. 🧹🔒

🎭 The Actor Model (Deep Dive)

While invoke is perfect for calling a single function, the Actor Model is for when you need to manage an entire, long-living, stateful process as a child of your main machine. Actors are simply other state machine interpreters that are "spawned" and managed by a parent.

0.4.1 Upgrade – SyncInterpreter Spawn: Synchronous parents can now spawn child actors without blocking. Under the hood, each child SyncInterpreter runs in its own threading.Thread, while the parent keeps processing events.

This pattern is the key to unlocking massive architectural freedom and managing complex concurrency with grace. It is fully supported in both the Interpreter (async) and SyncInterpreter (sync).

📐 Reference Diagram

The relationship is simple: a parent spawns a child, can send it messages (events), and the child can send messages back to its parent.

┌───────── Parent Machine ─────────┐
│                                  │
│  Action: "spawn_myActor"         │
│      │                           │
│      ▼                           │
│  ╔═════════════╗ Parent can send │
│  ║ Child Actor ║────────────────►│
│  ║ Interpreter ║◄────────────────┤
│  ╚═════════════╝ Child can send  │
│                  (via i.parent)  │
│                                  │
└──────────────────────────────────┘

🚀 spawn_ Actions — How Actors Are Born

Spawning a child machine is as simple as defining an action whose type starts with spawn_ (or spawn_blocking_).

Action Prefix	Async Interpreter	Sync SyncInterpreter	Use When…
spawn_<name>	Non‑blocking (child runs in a task)	Non‑blocking (child runs in a background thread)	Parent must stay responsive
spawn_blocking_<name>	Still non‑blocking (same as above)	Blocking (runs inline, parent waits)	You want strict in‑order execution

Behind the scenes the interpreter does four deterministic steps (code in Interpreter._spawn_actor / SyncInterpreter._spawn_actor):

1. Name resolution – strip the prefix to get <name> and look it up in machine.logic.services["<name>"].
2. Source validation
   • If it’s a MachineNode → use as‑is.
   • If it’s a callable → treat as a factory; call it with (interpreter, ctx, event) and expect a MachineNode back.
   • Otherwise → raise ActorSpawningError.
3. ID assignment – build a unique child ID: "{parentId}:{name}:{uuid4()}" (e.g. cart:paymentActor:3e3b6b9c-…).
4. Start & register – create a new interpreter for the child, start it, stash it in parent._actors[childId].

# ✅ Happy path: provide MachineNodes or factories in services
services = {
    "paymentActor": create_machine(payment_cfg, logic=payment_logic),

    # Factory variant – can use runtime data
    "dynamicActor": lambda i, ctx, e: create_machine(build_cfg_from_context(ctx))
}

# ❌ Wrong: will raise ActorSpawningError
services = {
    "oops": "not a machine"
}

📝 Remember: Spawned actors live until you stop them (or their parent stops). They inherit the parent’s plugins automatically, so logging/metrics stay consistent.

Messaging Patterns

The actor model enables powerful, interpreter-agnostic communication patterns.

Pattern	How It Works	Use Case
Request/Reply	A child actor performs a task and sends a RESULT or FAILURE event back to its parent when it completes.	warehouseRobot spawns a pathfinder actor to calculate a route and report back.
Command	The parent looks up a specific child in its interpreter._actors registry and sends it a command event like PAUSE.	A mediaPlayer machine tells its spawned volumeControl actor to mute.
Broadcast	The parent iterates over all interpreter._actors.values() and sends the same event to every child.	A collaborativeEditor machine tells all cursor actors to change color.
Escalation	A child actor hits an error and bubbles it up by calling i.parent.send("CHILD_FAILED", ...).	A low-level network actor fails, telling the app to show a global “Offline” banner.

▶️ Event Flow of an Actor Interaction

Let's trace the warehouseRobot example to see how the parent and child communicate:

1. Parent (warehouseRobot) enters the planning_route state.
2. Parent's entry action (spawn_pathfinder_actor) is executed. An Interpreter for the pathfinder machine is created and started. This is the Child Actor.
3. Child (pathfinder) immediately enters its calculating state.
4. Child invokes its calculate_path_service.
5. (...time passes...)
6. Child's service finishes successfully, triggering its onDone transition.
7. Child's onDone action (send_path_to_parent) is executed. Inside this action, it calls i.parent.send("PATH_CALCULATED", ...).
8. Parent's event loop receives the PATH_CALCULATED event.
9. Parent follows its own transition for this event, running the store_path action and moving to the moving state.
10. Child moves to its finished state and terminates.

This clear, decoupled communication is what makes the Actor Model so powerful for complex systems.

Need a refresher later? Jump back here: see Event Flow of an Actor Interaction.

Dynamic Pools & Supervision Strategies (Sync & Async)

You can spin up and tear down many child interpreters at runtime (worker pools, per-user sessions, etc.) and supervise them with a single, consistent pattern.

Spawning Workers Dynamically

Goal	Async Interpreter	Sync SyncInterpreter
Create a child machine on demand	Use an async-friendly action; child runs in its own asyncio task	Use a normal action; child runs in a background threading.Thread (parent .send() still blocks until the current loop settles)

# Shared worker config/logic assumed
worker_machine = create_machine(worker_cfg, logic=worker_logic)

# -------- ASYNC parent --------
async def spawn_worker(i: Interpreter, ctx, e, a):
    wid = ctx.setdefault("next_id", 0) + 1
    ctx["next_id"] = wid
    child = Interpreter(worker_machine, parent=i).start()  # await .start()
    i._actors[f"worker:{wid}"] = await child
    # Optionally send initial event
    await child.send("BOOTSTRAP", wid=wid)

# -------- SYNC parent --------
def spawn_worker(i: SyncInterpreter, ctx, e, a):
    wid = ctx.setdefault("next_id", 0) + 1
    ctx["next_id"] = wid
    child = SyncInterpreter(worker_machine).start()  # runs in its own thread internally
    i._actors[f"worker:{wid}"] = child
    # Initial command:
    child.send("BOOTSTRAP", wid=wid)

Supervision Strategies (Unified)

If a child crashes or its invoked service fails, have the child report up and the parent decide what to do.

Step	Child does…	Parent handles…
1	Add an onError on the critical invoke (or catch exceptions in actions)	Provide a global/state on: { "CHILD_FAILED": … } handler
2	In the error action, send a custom event to i.parent	Log, restart (spawn_* again), or escalate

Child error action (async vs sync):

# Async child
async def report_failure_to_parent(i: Interpreter, ctx, e, a):
    await i.parent.send(
        "CHILD_FAILED",
        actorId=i.id,
        error=str(e.data)
    )

# Sync child
def report_failure_to_parent(i: SyncInterpreter, ctx, e, a):
    i.parent.send(
        "CHILD_FAILED",
        actorId=i.id,
        error=str(e.data)
    )

Parent handler (same JSON for both):

"running_children": {
  "entry": "spawn_worker",
  "on": {
    "CHILD_FAILED": {
      "actions": ["log_child_error", "spawn_worker"],
      "target": ".running_children"
    }
  }
}

Tip: If you don’t want automatic restart, drop the target and just log/alert.

---

Debugging Actors

The built-in LoggingInspector automatically prefixes logs with the unique actorId of the interpreter, making it easy to trace concurrent operations.

parent_interpreter.get_snapshot() recursively includes the snapshots of all its child actors, giving you a complete picture of the entire system state.

You can test a child actor in complete isolation by creating an interpreter for it directly in your test suite, simulating parent messages by calling .send().

---

🏗️ Architectural Patterns — Extended

Automatic Logic Discovery — Under the Hood

When you use logic_modules or logic_providers, the LogicLoader performs these steps:

• Introspect: Scans provided modules/objects for all public callables (functions or methods).
• Normalize: Creates a lookup map with both original and camelCase names for each callable.
• Validate: Checks that each action/guard/service name in the JSON config exists in the lookup map, otherwise raises ImplementationMissingError.
• Bind: Constructs the final MachineLogic object for the interpreter to use.

The collision rule is simple: the last provider wins. If you have logic_providers=[ProviderA(), ProviderB()] and both have a do_task method, ProviderB's method will be used.

Hybrid Execution: The Best of Both Worlds

What if your application is mostly asynchronous, but you need to run a quick, blocking, deterministic check and get an immediate result? This library uniquely supports this hybrid model. An async parent machine can programmatically create, run, and get the result from a sync child machine within a single, atomic action.

Step 1: The Asynchronous Parent

// examples/hybrid/manufacturing_line/manufacturing_line_async.json
{
  "context": {
    "partId": null,
    "qcResult": null
  },
  "id": "assemblyLine",
  "initial": "acceptingParts",
  "states": {
    "acceptingParts": {
      "on": {
        "PART_RECEIVED": {
          "actions": "assign_part_id",
          "target": "assembly"
        }
      }
    },
    "assembly": {
      "invoke": {
        "src": "assemble_part",
        "onDone": "qualityControl",
        "onError": "failed"
      }
    },
    "qualityControl": {
      "entry": "run_quality_check",
      "on": {
        "QC_PASSED": "packaging",
        "QC_FAILED": "failed"
      }
    },
    "packaging": {
      "invoke": {
        "src": "package_part",
        "onDone": "complete",
        "onError": "failed"
      }
    },
    "complete": {
      "type": "final"
    },
    "failed": {
      "type": "final"
    }
  }
}

Step 2: The Synchronous Child

// examples/hybrid/manufacturing_line/quality_check_sync.json
{
  "context": {
    "inspectionCount": 0,
    "isPassed": false
  },
  "id": "qualityCheck",
  "initial": "inspecting",
  "states": {
    "inspecting": {
      "entry": "run_inspection",
      "on": {
        "": [
          {
            "guard": "did_pass",
            "target": "passed"
          },
          {
            "target": "failed"
          }
        ]
      }
    },
    "passed": {
      "entry": "set_passed"
    },
    "failed": {}
  }
}

Step 3: The Orchestration Logic

# examples/hybrid/manufacturing_line/hybrid_manufacturing.py
# ... (imports and other logic functions) ...

# 🚀 Hybrid Orchestration Action
def run_quality_check_action(
    interpreter: Interpreter,  # This is the PARENT async interpreter
    context: Dict,
    event: Event,
    action_def: ActionDefinition,
) -> None:
    """
    An action on the ASYNC machine that creates and runs the SYNC QC machine.
    """
    logging.info("----------")
    logging.info("🔍 Entering Quality Control. Running synchronous QC check...")

    # 1. Load the synchronous machine's config.
    with open("quality_check_sync.json", "r") as f:
        qc_config = json.load(f)

    # 2. Define the logic for the synchronous machine.
    qc_logic = MachineLogic(
        actions={
            "run_inspection": run_inspection_action,
            "set_passed": set_passed_action,
        },
        guards={"did_pass": did_pass_guard},
    )

    # 3. Create and run the SYNC interpreter. The .start() method
    #    is blocking and runs the entire machine to its final state.
    qc_machine = create_machine(qc_config, logic=qc_logic)
    qc_interpreter = SyncInterpreter(qc_machine).start()

    # 4. Now that the sync machine has finished, inspect its final state.
    if "qualityCheck.passed" in qc_interpreter.current_state_ids:
        context["qcResult"] = "PASSED"
        asyncio.create_task(interpreter.send("QC_PASSED"))
    else:
        context["qcResult"] = "FAILED"
        asyncio.create_task(interpreter.send("QC_FAILED"))

    logging.info(f"🏁 Synchronous QC check complete. Result: {context['qcResult']}")
    logging.info("----------")

Global Logic Registration for Large Applications

In larger applications, you might have logic spread across many different files. Instead of importing and passing a list of modules to create_machine every time, you can register them once when your application starts up.

The LogicLoader is a singleton, meaning there's only one instance of it throughout your application's lifecycle. You can get this instance and register modules globally.

# In your application's main entry point (e.g., main.py or __init__.py)
from xstate_statemachine import LogicLoader
import my_app.user_logic
import my_app.payment_logic

# 1. Get the single global instance of the LogicLoader
loader = LogicLoader.get_instance()

# 2. Register all your logic modules once during startup
loader.register_logic_module(my_app.user_logic)
loader.register_logic_module(my_app.payment_logic)

# --- Later, in a different part of your application ---

# 3. Now, create_machine will automatically find the registered logic
#    without needing the logic_modules argument.
from xstate_statemachine import create_machine

machine_one = create_machine(user_config)
machine_two = create_machine(payment_config)

Choosing Your Logic Style

Your library supports two primary ways of organizing your logic: functional (standalone functions in modules) and class-based (methods on a class instance). Both work with either explicit binding or auto-discovery. Here’s how to choose:

Logic Style	Best For	Pros	Cons
Functional	Simpler machines, stateless logic, or promoting a pure-function style.	- Easy to test individual functions. - Encourages stateless, predictable logic.	- Can become disorganized if many unrelated functions are in one file. - Sharing state requires passing it through the context.
Class-Based	Complex machines with related logic, or when logic needs its own internal state.	- Excellent organization; groups related logic together (FileUploaderLogic). - Can manage its own state via self in addition to the machine's context.	- Slightly more boilerplate (class definition, __init__). - Can be overkill for very simple machines.

---

🔁 Sync vs Async — Under the Hood

Concern	Interpreter (asyncio)	SyncInterpreter (blocking)
Event Loop	Runs in a background asyncio.Task	A while loop inside the .send() method
after Timers	Non-blocking (asyncio.create_task)	Non-blocking (threading.Thread)
invoke Services	Non-blocking (asyncio.create_task)	Blocking (runs inline)
spawn Actors	Non-blocking (asyncio.create_task)	Non-blocking (threading.Thread) or Blocking (inline)

🚀 What’s New in 0.4.1 for the SyncInterpreter

• Non‑blocking actor spawning: spawn_* actions now create child SyncInterpreters in background threads, keeping the parent responsive.
• Multiple concurrent after timers per state: You can declare several delays in the same state; each gets its own handle and unique ID.
• Safer shutdown: .stop() now stops all children first, then cancels timers—no dangling threads, no stray events.

💡 You still get deterministic, blocking .send() semantics—the non‑blocking bits are hidden behind thread coordination.

Rule of Thumb:

Desktop / CLI → SyncInterpreter

SyncInterpreter is still “blocking” where it matters: send() (and start()/stop()) run the full transition loop synchronously and only return when all exit/entry actions, guards, transient ("") transitions and internal events have fully settled. The only things that run out‑of‑band are after timers and spawned child interpreters (each in its own thread). So when this README says “non‑blocking” for SyncInterpreter, it only refers to those background helpers—not to the main event-processing call.

Web / IoT / pipelines → Interpreter

⚠️ Features not Supported by SyncInterpreter

While the synchronous engine is highly capable, it enforces one hard constraint to guarantee predictable, blocking behavior. Any violation raises NotSupportedError instantly:

Attempted Feature	Exception Raised	Guarding Method
Async callables in actions or services (coroutines / async def)	NotSupportedError	SyncInterpreter._execute_actions & SyncInterpreter._invoke_service

🧘 Why so strict? The SyncInterpreter's core send method must finish its event processing loop before returning control to the caller. Awaiting a coroutine would break that guarantee. The hard error surfaces this design mismatch early, nudging you towards either the full async Interpreter or a refactoring to synchronous logic.

Migrating Sync→Async

1. Swap interpreter class from SyncInterpreter to Interpreter.
2. Make your main function async, and await all calls to .start(), .send(), and .stop().
3. Replace blocking calls like time.sleep() with await asyncio.sleep().

Threads & Processes

• Use loop.call_soon_threadsafe() for cross‑thread .send() to an Interpreter.
• For cross‑process, bridge with a message queue and .send() from the process.

📡 Cross‑thread .send() with SyncInterpreter

SyncInterpreter isn’t inherently thread‑safe. If another thread must trigger events, use a tiny queue/dispatcher pattern:

# owner thread ----------------------------
svc = SyncInterpreter(machine).start()
inbox: "queue.Queue[tuple[str, dict]]" = queue.Queue()

def pump():
    while running:
        try:
            etype, payload = inbox.get(timeout=0.05)
            svc.send(etype, **payload)          # all sends happen on THIS thread
        except queue.Empty:
            pass

dispatcher = threading.Thread(target=pump, daemon=True).start()

# worker thread ---------------------------
def do_work():
    inbox.put(("UPDATE", {"value": 42}))

Alternatives

• Wrap svc.send(...) with a threading.Lock if very occasional cross‑thread calls are fine and you can guarantee no re‑entrancy.
• Expose a small send_threadsafe() helper that just enqueues into the owner’s inbox.

There is no built‑in call_soon_threadsafe equivalent for the sync engine—use a queue or redesign so only one thread owns the interpreter.

---

🤖 CLI: Boilerplate Generation

🧰 What is xsm and why should you care?

xsm is the command‑line companion to this library. It turns JSON charts into runnable Python code and keeps your event‑driven workflow consistent.

🎯 Goals

• Zero boilerplate: generate action/guard/service stubs and a runnable interpreter in seconds.
• Enforce conventions: consistent file names, logger setup, async/sync signatures.
• Speed up iteration: tweak JSON, regen, run—repeat.
• Promote EDA by default: generated runners log events, not function calls; you see flows, not stacks.

🏃 Typical workflow

1. Design machine visually → export JSON
2. Run:  xsm gt path/to/machine.json
3. Fill in generated stubs (actions/guards/services)
4. python my_machine_runner.py   # see it run

🧪 Newbie-friendly example

xsm gt light_switch.json
# ➜ creates light_switch_logic.py + light_switch_runner.py
python light_switch_runner.py
# Logs every event/transition; edit stubs to add real behaviour

🖼️ How it fits in the big picture

flowchart LR
    A[Stately Editor / JSON] -->|xsm gt| B[Generated Logic & Runner]
    B --> C[Your Business Code]
    C --> D[Interpreter / SyncInterpreter]
    D --> E[Running App / Tests / CLI Demo]

Use xsm when you…

• start a new machine and don’t want to hand‑wire 20 function placeholders
• onboard juniors—give them stubs with correct signatures instead of “read the docs and pray”
• keep large teams consistent (same folder structure, same logger, same sample main())

Skip xsm when you…

• prefer hand-crafted code (fine for tiny machines)
• already built your own templates or scaffolding
• need dynamic runtime generation (use create_machine() directly)

TL;DR — xsm is your scaffolder + guard rail for event‑driven, state-charted Python.

The CLI went through a sizeable upgrade in 0.4.1:

• The main entry-point is now xsm (short & sweet).

Note: The legacy console entry point xstate-statemachine is deprecated and may be removed in a future release. New installs expose only xsm. If you still call the old name, pin to an older version or add a shell alias.

• Aliases everywhere: quicker typing with gt, -jp, -jc, -fc, etc.
• Hierarchy helper: pass a parent + children, or let the CLI guess and confirm interactively.
• Safer defaults: allow_abbrev=False, clearer error messages, overwrite prompts, reserved‑keyword fixes, robust path resolution in runners.

Basic Usage

# Single file, default options (async, class-based, 2 files)
xsm generate-template path/to/machine.json

# Short alias
xsm gt path/to/machine.json

---

🔤 Aliases & Short Flags

Long	Short	Description
generate-template	gt	Main code-gen subcommand
--json-parent	-jp	Parent machine JSON path
--json-child	-jc	Child/actor JSON path (repeatable)
--file-count	-fc	1 = single file, 2 = logic + runner
--async-mode	-am	yes / no
--loader	-l	yes / no
--style	-s	class / function
--output	-o	Output directory
--force	-f	Overwrite without prompt

Boolean flags accept: yes/no, true/false, 1/0 (case-insensitive).

---

👑 Hierarchical Machine Generation

You can feed multiple JSON files that represent a parent + child actors. Two ways:

1. Explicit:

xsm gt -jp parent.json -jc child1.json -jc child2.json

2. Heuristic + Interactive (just pass all files, let CLI guess):

xsm gt parent.json child1.json child2.json
# CLI prints:
#   1. ParentMachine (looks like parent, score: 3)
#   2. ChildOne      (looks like child, score: 0)
#   3. ChildTwo      (looks like child, score: 0)
# Is this correct? [Y/n]

If you answer “n”, you’ll get a prompt to pick the parent by number.

---

🗂 One File vs Two Files

• -fc 2 (default): Generates:

  • *_logic.py with all stubbed actions/guards/services.
  • *_runner.py with a ready-to-run interpreter and simulation loop.

• -fc 1: Everything in one file; imports de-duplicated, logger setup unified. Good for quick demos or tiny scripts.

---

🧵 Async vs Sync Templates

• -am yes (default): Async Interpreter, asyncio.sleep, await interpreter.send(...).
• -am no: Blocking SyncInterpreter, time.sleep, direct .send(...).

The generator ensures signature correctness: actions/services async only when needed, guards always sync.

---

✨ Smarter Runner Code (What changed)

• Logic binding: Always correct for your chosen style (class/function) and loader mode.
• Path resolution: The runner looks for your JSON next to itself, then in the parent directory—handier after moving files.
• Keyword conflicts: If your JSON names collide with Python keywords, they are auto-fixed (e.g., def → def_).

---

🛡️ Validation & Prompts

• No partial option matches: --sl won’t silently match --sleep. Typos crash early.
• Multiple --json-parent? Immediate error.
• Existing files? You’ll be asked “Overwrite? [Y/n]” unless you pass -f.

---

📦 Quick Examples

1. Async + Class + Auto Loader (common case)

xsm gt examples/traffic_light.json

Creates:

traffic_light_logic.py    # class TrafficLightLogic
traffic_light_runner.py   # async main(), LoggingInspector, sleep loop

2. Sync + Function Style + Explicit Binding

xsm gt login_form.json -am no -s function -l no

You’ll get function stubs and a MachineLogic mapping generated for you.

3. Single-file, Hierarchical

xsm gt -jp parent.json -jc child.json -fc 1 -o ./gen

Produces ./gen/parent_logic.py (combined). Parent & child event simulations show up in one script.

---

🧪 Testing the Generator

Because the generated code is deterministic, you can:

• Assert file existence and contents (headers, logger lines, imports).
• Run pytest on your generated package right away—nothing is “magic”.

That’s the CLI in 0.4.1—faster, stricter, and hierarchy-aware. 🚀

---

🐞 Debugging & Visualization

This section dives into:

1. LoggingInspector patterns (verbosity, custom format).
2. Writing plugins with detailed lifecycle hooks.
3. Snapshots for crash‑recovery & golden tests.
4. Auto‑diagrams (Mermaid & PlantUML).
5. REPL live‑tinkering with await interp.send(...).

When something goes sideways at 2 AM you need clarity, not guess-work. XState-StateMachine ships with an instrumentation layer that lets you inspect, log, snapshot and draw every heartbeat of your machine.

🪵 Built‑in Logging Infrastructure

Out‑of‑the‑box the package exposes a library‑safe logger named "xstate_statemachine" with a NullHandler already attached (xstate_statemachine/logger.py). That means no more “No handler found” spam in consumer apps. Simply configure logging once in your entry point and every module—plus all plugins—will follow suit:

import logging

# Your application bootstrap 🔧
logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s | %(levelname)-8s | %(name)s | %(message)s",
)

# From here on the library logs seamlessly
from xstate_statemachine import create_machine, Interpreter
from xstate_statemachine.logger import logger  # Convenience re‑export 🪵

logger.info("✅ Logging initialised!")

Since v0.4.1, the generator emits stricter, standardized logger.info(...) messages (wording fixed, emojis intact); if you grep/assert on logs, update expectations — core logging behavior is unchanged.

What	Where	Why
Package logger	logging.getLogger("xstate_statemachine")	Hierarchical—sub‑modules inherit handlers & level.
NullHandler pre‑installed	Added in the library to be polite	Prevents accidental stderr noise in apps that forget basicConfig().
Helper logger constant	from xstate_statemachine.logger import logger	Quick access for your actions / guards without calling getLogger() each time.

💡 Tip: Add multiple handlers (file, JSON, OTEL, …) in your own basicConfig or custom setup—the library will respect them automatically.

1. LoggingInspector Plugin 🕵️‍♀️

import logging
from xstate_statemachine import Interpreter, LoggingInspector

logging.basicConfig(level=logging.INFO, format="%(asctime)s | %(message)s")

machine  = create_machine(cfg)
service  = Interpreter(machine).use(LoggingInspector())   # 🔑 register _before_ .start()
await service.start()

Sample output (collapsed)

2025-07-10 16:02:15 | 🕵️ STATE  idle  ➡  loading  (on FETCH)
2025-07-10 16:02:15 | ⚙️  Action: setSpinner(True)
2025-07-10 16:02:15 | ☁️  Invoke: fetchData()
2025-07-10 16:02:16 | ✅ done.invoke.fetchData  — 200 OK
2025-07-10 16:02:16 | 🕵️ STATE  loading  ➡  success  (on done.invoke.fetchData)
2025-07-10 16:02:16 | ⚙️  Action: cacheData

🔧 Customising the log stream

class ShortLog(LoggingInspector):
    """Print a single‑line summary for every *external* transition."""

    def on_transition(self, interpreter, from_states, to_states, transition):
        old = ",".join(s.id for s in from_states)
        new = ",".join(s.id for s in to_states)
        print(f"[{transition.event}] {old} → {new}")

Attach multiple inspectors—analytics, tracing, etc.—each focusing on a single concern.

---

2. Writing Your Own Plugin 🔌

All plugins inherit from PluginBase and can tap into the interpreter’s lifecycle. Below are the stable hooks available today (everything else is considered experimental or on the roadmap).

🧩 Plugin Hook Matrix — copy‑paste ready 📝

🔗 Hook	Python Signature	📅 When it Fires	💡 Typical Use
🏁 on_interpreter_start	on_interpreter_start(self, interpreter)	Right after interpreter.start() begins	Initialise DB connections, timers, metrics
🛑 on_interpreter_stop	on_interpreter_stop(self, interpreter)	As soon as interpreter.stop() is invoked	Flush buffers, close sockets
✉️ on_event_received	on_event_received(self, interpreter, event)	Every time an event is dequeued for processing	Audit trails, event‑level analytics
🔀 on_transition	on_transition(self, interpreter, from_states, to_states, transition)	After states are exited → actions run → new states entered	Tracing, Prometheus counters, BI pipelines
⚙️ on_action_execute	on_action_execute(self, interpreter, action)	Immediately before an individual action implementation runs	Profiling, APM spans, debugging prints
🛡️ on_guard_evaluated	on_guard_evaluated(self, interpreter, guard_name, event, result)	After a guard function's condition is evaluated	Debugging guard logic, conditional analytics
📞 on_service_start	on_service_start(self, interpreter, invocation)	Just before an invoked service begins execution	Logging service calls, tracing external interactions
✅ on_service_done	on_service_done(self, interpreter, invocation, result)	When an invoked service returns successfully	Logging successful outcomes, result processing
💥 on_service_error	on_service_error(self, interpreter, invocation, error)	When an invoked service raises an exception	Error reporting, failure handling, alerting

🛠️ Tip: Implement only the hooks you need; methods left un‑overridden incur zero overhead thanks to Python’s dynamic dispatch. 🚀

from xstate_statemachine import PluginBase

class PromMetrics(PluginBase):
    """Increment a Prometheus counter on every state change."""

    def __init__(self, counter):
        self.counter = counter

    def on_transition(self, interpreter, from_states, to_states, transition):
        # Label by triggering event for easy dashboard filters
        self.counter.labels(event=transition.event).inc()

Just .use(PromMetrics(prom_counter)) on an interpreter instance and you’re collecting metrics! 📈

---

3. Snapshots 🧩

Persist the exact runtime status—active states and mutable context—to disk or Redis.

snap = interpreter.get_snapshot()    # JSON str
# Later — even after deploy
restored = await Interpreter.from_snapshot(snap, machine).start()

🔄 Restoring from a Snapshot — Mind the MachineNode 📂

A snapshot captures only dynamic runtime data:

1. status (running / stopped)
2. context dict
3. IDs of active states

It does not store the static state‑chart structure itself. Therefore Interpreter.from_snapshot() needs the original MachineNode— the same object returned by create_machine(...)—to rebuild the interpreter.

# ✅ Happy Path
machine  = create_machine(cfg, logic_modules=[app_logic])

service   = await Interpreter(machine).start()
snap      = service.get_snapshot()     # JSON string
await service.stop()

# later / after restart
restored  = await Interpreter.from_snapshot(snap, machine).start()

Note (async only): from_snapshot(...) exists on Interpreter right now. SyncInterpreter doesn’t have a restore helper yet—use create_machine(...), then manually set context and drive events to reach the desired states, or switch to the async interpreter for snapshot restore flows.

# 😬 Wrong – passing raw JSON
cfg = json.load(open("chart.json"))
snap = ...                       # previously saved

# Re‑creating a *different* MachineNode (new object)
machine2 = create_machine(cfg)   # ⚠️ distinct in memory

await Interpreter.from_snapshot(snap, machine2)  # ❌ Raises StateNotFoundError

🧩 Tip: Keep the original MachineNode in a module‑level variable, or persistently cache it, so restoration is trivial. Creating a byte‑for‑byte identical MachineNode works too, but re‑running create_machine() must use the exact same JSON + logic to avoid ID drift.

Use-cases:

• Resilient workers—crash-safe resume after process restarts
• Time-travel debugging—save before risky ops, restore in REPL
• CI golden tests—diff snapshots to detect un-intended behavioural drift

---

4. Auto-Generating Diagrams 🖼️

Keep docs evergreen by baking diagram generation into CI.

mermaid = machine.to_mermaid()
with open("docs/statechart.mmd", "w") as f:
    f.write("```mermaid\n" + mermaid + "\n```")

Or, for architects living in PlantUML:

plantuml = machine.to_plantuml()
Path("docs/diagram.puml").write_text(plantuml)

Integrate with mkdocs-material, GitHub Pages, Confluence—anything that renders Mermaid/PUML—your diagrams will always mirror the code running in prod.

📤 Programmatic Exports: machine.to_dict() & machine.to_json()

Besides Mermaid/PlantUML, you can export the static machine definition directly:

cfg_dict  = machine.to_dict()   # → plain Python dict (JSON‑serialisable)
cfg_json  = machine.to_json()   # → JSON string (pretty by default)
Path("docs/machine.json").write_text(cfg_json)

What’s included? The full, normalised XState JSON graph (ids, states, on/after/invoke, etc.).

What’s NOT included? Runtime data (context values, active states, timers). For that, use interpreter.get_snapshot() / Interpreter.from_snapshot(...).

Tip: to_dict() is ideal for diffing machines in tests or piping into other tooling; to_json() is handy for bundling with front‑end visualisers.

5. REPL Live‑Tinkering 💻

Why bother? • Instant feedback when wiring new actions or guards • Zero‑compile “what happens if…?” exploration • Perfect for smoke‑testing machines that talk to live APIs or devices

5.1  Pick a REPL with top‑level await

REPL	Setup	Remarks
IPython ≥ 8.0	pip install ipython ipython --autoawait asyncio	Rich tracebacks & tab‑completion
ptpython	pip install ptpython	Built‑in asyncio, syntax highlighting
Vanilla Python 3.12+	python -m asyncio	Stock interpreter now supports top‑level await 🎉

5.2  Bootstrap an interpreter session

# light_repl.py
import json, asyncio
from xstate_statemachine import create_machine, Interpreter, LoggingInspector
import light_switch_logic  # ← your actions

cfg     = json.load(open("light_switch.json"))
machine = create_machine(cfg, logic_modules=[light_switch_logic])

# Create the async interpreter but DON'T start the event loop yet
service = Interpreter(machine).use(LoggingInspector())

Launch IPython with the pre‑wired objects:

ipython --autoawait asyncio -i light_repl.py

5.3  Play!

In [1]: await service.start()
🕵️ STATE off

In [2]: await service.send("TOGGLE")
🕵️ STATE off ➡ on   (on TOGGLE)

In [3]: service.current_state_ids
Out[3]: {'lightSwitch.on'}

In [4]: service.context
Out[4]: {'flips': 1}

Tip — alias event sending to shorten typing:

In [5]: %alias send await service.send
In [6]: send TOGGLE

5.4  Hot‑reload without leaving the REPL

In [7]: %load_ext autoreload
In [8]: %autoreload 2   # picks up edits in light_switch_logic.py

# tweak your action, hit save, then...
In [9]: send TOGGLE     # new code runs immediately

5.5  Snapshot & rewind on the fly

In [10]: snap = service.get_snapshot()

# …experiment wildly…
In [11]: await service.send("GLITCH_EVENT")

# Restore pristine state
In [12]: from xstate_statemachine import Interpreter
In [13]: service = await Interpreter.from_snapshot(snap, machine).start()

5.6  Deep‑dive tricks

Trick	Command
Inspect queued events	service._event_queue.qsize()
Peek next transition	machine.get_next_state("lightSwitch.on", {"type": "TOGGLE"})
Pause timers in tests	await service.stop(); asyncio.get_running_loop().set_debug(False)
Spawn a child REPL for actors	child = next(iter(service._actors.values())); await child.send(...)

🚀 You now have an always‑on laboratory for your state machines—no rebuilds, no deployment cycles, just pure interactive discovery. Happy tinkering!

🔼 Upgrade Notes: 0.4.0 → 0.4.1

No breaking API removals, but a few behaviour tweaks to be aware of:

SyncInterpreter

• Actor spawning no longer blocks the parent. If you expected strict in-order execution, add explicit joins or guards.
• Multiple after timers per state can now fire; ensure your guards/actions handle overlapping timers if you add many.

CLI

• Command renamed to xsm. The legacy xstate-statemachine console entry point is deprecated and may disappear in a future release—migrate scripts to xsm.
• Interactive hierarchy guesser: If you pass multiple JSONs without --json-parent, you’ll see a prompt. Use -f to skip overwrites, not prompts.
• Strict arg parsing: Abbreviated flags no longer work; typo’d flags now error (good!).

Generated Code

• Keyword auto-fix: If an action/guard/service name is a Python keyword, it’s renamed (e.g., class → class_). Search for # alias for JSON name comments.
• Runner path resolution: Scripts now try two locations for the JSON. If you were relying on Path.cwd(), adjust accordingly.

If you used any internal/private APIs, re-run your tests—internals around threading and timer bookkeeping have changed for robustness.

---

📑 API Reference

Below is the complete, in-depth guide to the library's public API. This section details every key class, function, and exception, with explanations and usage examples to help you master the library.

---

1. The Factory: create_machine

This is the primary, user-facing entry point for creating a state machine instance. Its job is to take your declarative JSON config and your Python business logic and weave them together into a single, executable MachineNode object.

Signature:

create_machine(
    config: Dict[str, Any],
    *,
    logic: Optional[MachineLogic] = None,
    logic_modules: Optional[List[Union[str, ModuleType]]] = None,
    logic_providers: Optional[List[Any]] = None
) -> MachineNode

Parameters In-Depth

• config: Dict[str, Any] | Required The Python dictionary parsed from your state machine's JSON definition. Example:

  import json
  
  with open("my_machine.json") as f:
      machine_config = json.load(f)
  
  machine = create_machine(machine_config, ...)
  

• logic: Optional[MachineLogic] | Explicit Binding Explicitly map every action, guard, and service name to Python functions:

  from my_logic_file import my_action_func, my_guard_func
  
  explicit_logic = MachineLogic(
      actions={"doSomething": my_action_func},
      guards={"canDoSomething": my_guard_func}
  )
  
  machine = create_machine(config, logic=explicit_logic)
  

• logic_modules: Optional[List[Union[str, ModuleType]]] | Functional Auto-Discovery Auto-discover standalone functions in modules:

  import my_app.logic.user_actions
  
  machine = create_machine(config, logic_modules=[my_app.logic.user_actions])
  

• logic_providers: Optional[List[Any]] | Class-Based Auto-Discovery Auto-discover methods in class instances:

  from my_logic_file import BusinessLogicHandler
  
  logic_handler = BusinessLogicHandler()
  machine = create_machine(config, logic_providers=[logic_handler])
  

---

2. The Interpreters: Interpreter & SyncInterpreter

These are the engines that run your machine. They take a MachineNode (created by create_machine), manage its state, process events, and execute logic.

Properties

Property	Type	Description
.current_state_ids	Set[str]	All currently active state IDs. Useful for parallel states.
.context	Dict	The live, mutable context (memory) of the running machine instance.
.status	str	Lifecycle status: "uninitialized", "running", or "stopped".

Methods

Method	Returns	Description
.start()	self	Starts the interpreter, enters the initial state, and runs entry actions. Must be awaited for Interpreter.
.stop()	None	Stops the interpreter. For Interpreter, it cancels all running tasks. Must be awaited for Interpreter.
.send(event, **payload)	None	Sends an event to the machine. For Interpreter, this is an async operation.
.use(plugin)	self	Registers a plugin. Must be called before .start().
.get_snapshot()	str	Returns a JSON string of the current status, context, and active state_ids.
.from_snapshot(snap, machine)	Interpreter	(Class Method) Restores an interpreter from a saved snapshot.

_{† from_snapshot is available on Interpreter only (async). The sync engine currently has no equivalent helper.}

Usage Examples

# Async Interpreter Start
interpreter = await Interpreter(machine).start()

# SyncInterpreter Start
interpreter = SyncInterpreter(machine).start()

# Stop Interpreter
await interpreter.stop()

# Send events
await interpreter.send("UPDATE_USER", name="Alice", id=123)
await interpreter.send({"type": "SUBMIT"})

# Use plugins
interpreter.use(LoggingInspector()).use(MyCustomPlugin())

# Snapshot and restore
saved_state = interpreter.get_snapshot()
restored_interp = Interpreter.from_snapshot(saved_state, machine)

---

3. Core Logic & Model Classes

🔍 Handy MachineNode Helper Methods

When writing white‑box tests, REPL experiments, or CLI tools you often need to poke the state tree without spinning up a full interpreter. Two small but mighty helpers live right on the MachineNode:

Method	Returns	What it does	Typical Use‑Case
machine.get_state_by_id(state_id)	StateNode | None	Deep‑searches the tree for an exact, fully‑qualified state ID.	Assert a specific node exists, fetch its metadata in tests
machine.get_next_state(from_state_id, event)	Set[str] | None	Pure function that calculates where the machine would go from a given leaf state if event were sent. Note: Guards are ignored.	Fast unit tests for transition maps, generating coverage matrices

from xstate_statemachine import create_machine, Event

# Assume light_config is loaded from your JSON
light_config = {
    "id": "light",
    "initial": "green",
    "states": {
        "green": {"on": {"TIMER": "yellow"}},
        "yellow": {}
    }
}

# 1. Get the MachineNode instance
machine = create_machine(light_config)

# 2. Look up a specific node object
green_node = machine.get_state_by_id("light.green")
print(f"Found node: {green_node.id}")

# 3. Predict the outcome of an event without an interpreter
next_states = machine.get_next_state("light.green", Event(type="TIMER"))
assert next_states == {"light.yellow"}
print(f"From 'green', a 'TIMER' event would transition to: {next_states}")

💡 Tip: Because get_next_state is side‑effect‑free, you can call it in tight loops to generate all reachable paths for property‑based testing. Pair it with Hypothesis or pytest‑cases for powerful graph validation! 🧪✨

Class	Description
MachineNode	The parsed, in-memory representation of your entire state machine graph. Returned by create_machine, it's the central object passed to interpreters and used for static analysis
MachineLogic	Container for explicit action, guard, and service bindings.
Event	NamedTuple(type, payload) for events sent to the machine.
ActionDefinition	Represents a configured action from JSON, including .params for static data.

---

4. Extensibility & Debugging

Class	Description
PluginBase	Abstract base for custom plugins (override on_* methods).
LoggingInspector	Built-in plugin for detailed console output of state transitions, actions, and events.

---

5. Exception Hierarchy

All custom exceptions inherit from XStateMachineError.

XStateMachineError
 ├── InvalidConfigError
 ├── StateNotFoundError
 ├── ImplementationMissingError
 ├── ActorSpawningError
 └── NotSupportedError

• InvalidConfigError: Invalid machine JSON (e.g., missing id or states).
• StateNotFoundError: Transition target ID not found in the machine.
• ImplementationMissingError: Missing action, guard, or service implementation.
• ActorSpawningError: Error spawning a child actor/service.
• NotSupportedError: Using an async def function as an action or service with SyncInterpreter.

Use them in pytest.raises() to assert mis-configurations early.

---

🧬 Advanced Concepts

1. Supervision Trees 🌲 (Actor Failure Handling)

• Children may crash (uncaught exception in action/service).
• Parent decides policy:
  • onError transition → recover / restart actor.
  • Escalate — re-raise and crash upstream (default).
  • Silent — ignore by having no handler (not recommended).

2. Dynamic Spawn (PoC Micro-services)

Spawn machinery from runtime data:

async def spawn_tenant_actor(i, ctx, e, a):
    tenant_id = e.payload["id"]
    cfg = await fetch_tenant_machine(tenant_id)
    logic = await load_tenant_logic(tenant_id)
    return create_machine(cfg, logic=logic)

3. Hot Reload in Development ♻️

• Detect file change via watchdog.
• .stop() the old interpreter.
• Re-create_machine() with new JSON.
• .start(snapshot) to keep context.

Enjoy live editing of statecharts without losing session data.

4. Performance Tuning

Technique	When to use	Effect
Disable event‑loop debug mode asyncio.get_running_loop().set_debug(False)	After you’ve ironed out the bugs and want maximum throughput in production	Removes costly asyncio debug assertions (≈ 5‑10 % speed‑up in micro‑benchmarks)
Prefer SyncInterpreter for synchronous workflows	CLI tools, deterministic unit tests, or CPU-bound pipelines where asyncio is not needed.	Zero coroutine overhead, significantly faster per event in tight loops.
Bulk‑fire events API (planned)	High‑volume telemetry or log ingestion	Will let you enqueue a list of events in one syscall, minimising context‑switches

5. Architectural Pattern: invoke vs. async Actions

Your library offers two powerful ways to handle asynchronous operations, and choosing the right one is key to clean architecture.

Use invoke for True Asynchronous "States"

invoke is best when the machine enters a state that is defined by the running of an async task. Think of a loading state—its entire purpose is to wait for a service to complete.

• Declarative: Success (onDone) and failure (onError) are part of the state's definition, making the flow extremely clear from the diagram.
• Automatic Cleanup: If the machine transitions away from the "invoking" state, the library automatically cancels the running service task for you. This prevents orphaned tasks and race conditions.
• Use Case: API calls, database queries, or any task where you have distinct success and failure outcomes that lead to different states.

The api_fetcher example is a perfect illustration of this pattern.

"loading": {
  "invoke": {
    "src": "fetch_user_from_api",
    "onDone": {
      "target": "success",
      "actions": "set_user_data"
    },
    "onError": {
      "target": "failure",
      "actions": "set_error_message"
    }
  }
}

Use async Actions for "Fire and Forget" Logic

Sometimes, an async task is just a side effect, not the entire purpose of the state. It might have multiple, complex outcomes beyond simple success/failure, or it might not need to be cancelled if the state changes.

• Imperative Control: An async action gives you full control. It can send multiple different events back to the machine at different times to signal various outcomes.
• No Automatic Cleanup: The library will not automatically cancel an async action if the state changes. This can be useful for tasks that should complete regardless (like logging), but requires manual management for tasks that shouldn't.
• Use Case: Tasks that trigger other complex workflows, operations with more than two outcomes, or side effects that don't define the current state.

The data_fetcher example demonstrates this pattern beautifully. The fetch_data_action is just a task that runs when the fetching state is entered. It is responsible for sending FETCH_SUCCESS or FETCH_FAILURE back to its own interpreter to drive the next state transition.

# From data_fetcher_logic.py
async def fetch_data_action(
    self, i: Interpreter, ctx: Dict, e: Event, a: ActionDefinition
):
    # ... logic to fetch data ...
    if successful:
        # Manually send the success event
        await i.send("FETCH_SUCCESS")
    else:
        # Manually send the failure event
        await i.send("FETCH_FAILURE")

When to Use...

Aspect	Use invoke	Use an async Action
Clarity	When the flow is a clear success/failure branch.	When you have multiple, custom outcomes.
Cancellation	When the task must be cancelled on state exit.	When the task can run to completion regardless.
Simplicity	For most common async needs (API calls, etc.).	For more complex, imperative orchestrations.
Data Flow	Result is automatically passed to onDone/onError via event.data.	You must manually construct and .send() events with payloads.
Best Fit	A loading state whose entire purpose is the async call.	A "fire-and-forget" side effect within a broader state.

By understanding this distinction, you can model your asynchronous logic with even greater precision and clarity.

---

🌟 Best Practices

✅ Do	🚫 Avoid	Why
Keep state graphs small & composable	Monolithic 500-node monsters	Easier mental model; actors > beasts
Store quantitative data in context	Encoding counts/arrays in state IDs	Context is for numbers & strings; IDs are for qualitative phases
Use guards for business rules	Packing if logic inside actions	Guards are deterministic; actions are side-effects
Prefer after timers	asyncio.create_task(sleep()) inside actions	Declarative ≠ spaghetti
Model failures explicitly (error, timeout)	Relying on try/except deep inside services	Errors become testable & visible in diagrams
Name events imperatively (FETCH_USER)	Vague names (DO_IT, NEXT)	Better logs, clearer arrows
Unit-test machines head-less	UI-driven tests only	Faster CI; assert pure behaviour
Snapshot critical flows in CI	Trusting human QA memory	Catch regressions at graph-level
Document with Mermaid auto-build	Manually exported PNGs	Zero-drift diagrams

• Treat spawn_* like any other action name — the loader no longer special-cases them. Keep the naming clear: spawnPaymentActor, spawn_worker, etc.

Naming Conventions

• Events: SCREAMING_SNAKE_CASE
• States: camelCase preferred (loadingData)
• Action / Guard / Service names:
  • Python snake_case ↔ JSON camelCase (auto-mapped)
  • Prefix actors with a verb: spawnPaymentActor

File Layout (Suggestion)

myapp/
 ├─ statecharts/
 │   ├─ user_signup.json
 │   └─ payments/
 │       ├─ payment_flow.json
 │       └─ refund_actor.json
 ├─ logic/
 │   ├─ signup_logic.py
 │   └─ payments/
 │       └─ payment_logic.py
 └─ runners/
     └─ simulate_signup.py

Testing Tips 🧪

import pytest, json
from xstate_statemachine import create_machine, SyncInterpreter

@pytest.fixture
def signup_machine():
    cfg = json.load(open("statecharts/user_signup.json"))
    m   = create_machine(cfg, logic_modules=[signup_logic])
    return SyncInterpreter(m).start()

def test_happy_path(signup_machine):
    signup_machine.send("START")
    signup_machine.send("VALID_EMAIL")
    assert "signup.success" in signup_machine.current_state_ids

• Use SyncInterpreter even for async machines in unit tests – by stubbing async services as sync fakes.
• Compare snapshots instead of deep context asserts if the shape is large.

Unit-Testing Transitions with get_next_state()

For more granular testing, the MachineNode object includes a powerful utility, .get_next_state(from_state_id, event), for validating your machine's flow without running a full interpreter. It's a pure function that calculates the result of a transition without executing any actions or guards, making it perfect for fast unit tests.

from xstate_statemachine import Event

def test_timer_transition(light_machine):
    # light_machine is the MachineNode instance created by create_machine()
    event = Event("TIMER")

    # Calculate the expected next state without running actions
    next_states = light_machine.get_next_state("light.green", event)

    assert next_states == {"light.yellow"}

🎨 Style Guide (for actions/guards/services)

1. Always type-hint every arg & return.
2. Actions mutate ctx only; never sleep.
3. Guards are pure, side-effect-free, < 20 LOC.
4. Services should raise domain errors, not swallow them.
5. Log at source:

   logger = logging.getLogger("machine.payments")
   logger.info("Charging card %s...", ctx["card_id"])

6. Use emoji prefixes in logs for quick grep (consistent across repo).

---

❓ FAQ

Question	Answer
Is this library production ready?	Yes. It powers real-time IoT gateways handling 50k msgs/min and multiple SaaS dashboards.
Can I edit the JSON at runtime?	Absolutely. Re-create_machine() + .start(snapshot) to hot-swap.
Does it support PyPy?	✅ PyPy 3.10 passes the full test-suite.
How is it different from transitions?	XState-StateMachine implements full statecharts (hierarchy, parallelism, invoke, actors) and consumes XState JSON, not imperative decorators.
Can I use Pydantic in context?	Yep—store a model instance; just remember context is shallow-copied on interpreter start.
Where is the GUI inspector?	On the roadmap. Use Stately web simulator + LoggingInspector meanwhile.
Is there a code-gen for Python from Stately?	Not needed—export JSON → run. Zero translation.
Why did the CLI command change to xsm?	To make everyday usage faster and cleaner. It’s still the same project—just a friendlier binary.
Can the SyncInterpreter now spawn actors without blocking?	Yes! As of 0.4.1, children run in background threads. The parent keeps processing events synchronously.
How many after timers can I declare in one state (Sync)?	As many as you need. Each gets its own handle; all are cancelled on state exit.
The generator renamed one of my functions to if_ — why?	Your JSON used a Python keyword. 0.4.1 auto-appends _ to keep code valid. The original JSON name is kept as an alias.

---

🗺️ Roadmap (Next Release)

• SyncInterpreter.from_snapshot(...) helper – parity with async restore; single-call reconstruction from saved snapshots.
• Unified cancellation primitives – expose a consistent API over asyncio.Task/threading.Thread so plugins can introspect both.
• Bulk event enqueue API – send a list of events in one call for high‑volume telemetry pipelines.
• GUI Inspector prototype – lightweight web panel to watch events, guards, and context diffs live.

🤝 Contributing

Contributions are welcome and greatly appreciated! This project thrives on community involvement, and we're excited to see what you'll bring.

We follow a standard "Fork & Pull Request" workflow. Before submitting, please ensure your changes are well-tested, formatted correctly, and documented in the changelog.

For a full, step-by-step guide on how to:

• Set up your development environment
• Run tests and linting checks
• Submit your changes

Please see our Contributing Guide.

📜 License

MIT. In short:

Copyright (c) 2025 Basil T T

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the “Software”), to deal
in the Software without restriction…

See LICENSE file for the full legalese.

---

🎉 Congratulations! You’ve reached the end of the XState-StateMachine saga. Go forth and build bug-proof, self-documenting workflows! 🚀