From a Python function

"""
@Transformation.from_function — schema inference + idempotent upsert.

Input types come from parameter annotations; output keys come from the return
annotation (TypedDict, literal return dict, or a single `result`). Re-running
the same source is a no-op; changing it triggers an update.
"""

from typing_extensions import TypedDict

from vectorshift import Transformation

# 1. Scalar return → a single `result` output, inferred as int32.
@Transformation.from_function
def add(a: int, b: int) -> int:
    """Adds two integers."""
    return {"result": a + b}

print(f"1. add: id={add.id}, outputs={list(add.outputs)}")
print(f"   run: {add.run(inputs={'a': 2, 'b': 3})['outputs']}")

# 2. Multi-output via a TypedDict return → one output per key.
class Stats(TypedDict):
    total: int
    count: int

@Transformation.from_function(name="sum_and_count")
def stats(values: list) -> Stats:
    """Sum a list and report how many items it had."""
    return {"total": sum(values), "count": len(values)}

print(f"2. stats: outputs={list(stats.outputs)}")

# 3. defer=True → build locally, save when ready.
@Transformation.from_function(defer=True)
def shout(text: str) -> str:
    return {"result": text.upper()}

print(f"3. shout (deferred): id={shout.id}")   # None until saved
shout.save()
print(f"   saved: id={shout.id}")

# 4. Idempotency — re-decorating unchanged source updates nothing.
@Transformation.from_function
def add(a: int, b: int) -> int:
    """Adds two integers."""
    return {"result": a + b}

print(f"4. re-decorated add: same id={add.id}")

1. add: id=..., outputs=['result']
   run: {'result': 5}
2. stats: outputs=['total', 'count']
3. shout (deferred): id=None
   saved: id=...
4. re-decorated add: same id=...