Title: Python 3.14 Compatibility: NameError in _outbound_schedule_activity due to eager closure-cell inspection

What are you really trying to do?

I am upgrading a production service to use the latest Python environments (specifically testing against Python 3.14 alpha/dev builds). I encountered a regression where Temporal workflows fail immediately when scheduling activities due to changes in how Python 3.14 handles asyncio.Task initialization and closure cell inspection. I am reporting this so the SDK can be made compatible with the upcoming Python 3.14 release.

Describe the bug

In Python 3.14, any workflow calling await workflow.execute_activity(...) fails with a NameError.

This is caused by a "binding lag" in _workflow_instance.py. The variable handle is annotated but not assigned a value before it is referenced inside the run_activity closure. While previous Python versions (up to 3.13) deferred the inspection of these cells until the coroutine actually executed, Python 3.14’s asyncio.Task constructor (specifically when set_name is called) eagerly inspects the cells. Because the task is created in the same expression that binds the name, the inspection happens while the cell is still empty, triggering the error.

Traceback:

  File ".../temporalio/worker/_workflow_instance.py", line 1921, in _outbound_schedule_activity
    handle = _ActivityHandle(self, input, run_activity())
  File ".../temporalio/worker/_workflow_instance.py", line 3047, in _ActivityHandle.__init__
    instance._register_task(self, name=f"activity: {input.activity}")
  File ".../temporalio/worker/_workflow_instance.py", line 2458, in _register_task
    task.set_name(name)
NameError: cannot access free variable 'handle' where it is not associated with a value in enclosing scope

Minimal Reproduction

1. Use a Python 3.14 environment (e.g., cgr.dev/chainguard/python:latest-dev).
2. Run any standard "Hello World" workflow.
3. Attempt to execute an activity.
4. The worker will catch a NameError during the _outbound_schedule_activity call.

The problematic code pattern in _workflow_instance.py:

handle: _ActivityHandle  # Created but UNBOUND

async def run_activity() -> Any:
    while True:
        handle._started = True  # Closure reference
        ...

# Python 3.14 fails here because run_activity() 
# is inspected during Task creation inside the _ActivityHandle init
handle = _ActivityHandle(self, input, run_activity()) 

Environment/Versions

• OS and processor: Linux (verified on x86_64/Aarch64 via Chainguard images)
• Temporal Version: SDK versions 1.21.1 through 1.26.0 (current stable range)
• Python Version: 3.14.0a (early 2026 dev builds)
• Deployment: Docker-based workers

Additional context

A temporary runtime patch confirms that pre-binding the variable resolves the issue:

handle: Optional[_ActivityHandle] = None # Explicit binding fixes the NameError

This issue will likely affect all users as they begin migrating to Python 3.14 later this year, or those using "floating" rolling-release base images in their CI/CD pipelines.

Start in temporalio/worker/_workflow_instance.py at _outbound_schedule_activity and inspect how handle is captured by run_activity during _ActivityHandle construction. Reproduce the minimal Hello World activity workflow on a Python 3.14 development environment, then verify that scheduling the activity completes without the reported NameError.