[Bug] workflow.start_child_workflow() behaving differently than workflow.execute_child_workflow() in tests

[cpg1111]

What are you really trying to do?

Hi there, I am writing tests for a workflow in which it spawns N child workflows using workflow.start_child_workflow(), places them in a list, and resolves the result later, like the following example:

class Child:
    async def run(self) -> bool:
        return True

class Parent:
    async def run(self, n: int):
        children = []

        for _ in range(n):
            child = await workflow.start_child_workflow(Child, ...)
            children.append(child)

        ...
        for child in children:
            result = await child.result()
            # do something with result 

and test code like the following:

async def test_parent():
    # setup test env worker with both workflows
    wf = await env.client.start_workflow("parent", ...)
    await env.sleep(duration=timedelta(seconds=5))
    await wf.result()

Describe the bug

Even when I have that exact child workflow, I get an InvalidStateError saying Result is not set. If I do not put the handler in a list and immediately call await child.result() it blocks indefinitely. However if I use workflow.execute_child_workflow() everything works fine. It's also worth noting these workflows work fine in an actual execution, this issue only exists using the testing.WorkflowEnvironment.

Minimal Reproduction

Write a workflow that calls multiple child workflows using workflow.start_child_workflow(), write a test with a single sleep before awaiting the parent's result.

Environment/Versions

• Ubuntu 22.04 AMD EPYC 7352
• Temporal 1.22.5 and Python SDK 1.3.0
• from source

Additional context

N/A