[IzaakGough]

A Pub/Sub message that has attributes but no data crashed on_message_published with TypeError: Message.__init__() missing 1 required positional argument: 'data' before the handler ran.

This lets those messages through with data set to None, makes None the default for data when building a Message directly, and types publish_time as the datetime it already is.

• ordering_key is now str | None = None. The field after data needs a default too, and the handler already passed None when there is no ordering key.
• data is now typed str | None, so type checkers will flag code that passes msg.data straight to b64decode. The emulator sends "" for attribute-only messages, so the handler turns that into None too.

Tested with unit tests and in the emulator with an attribute-only message. Not tested with a real Pub/Sub delivery.

[gemini-code-assist]

Code Review

This pull request updates the Message class in pubsub_fn.py to change the type of publish_time to a timezone-aware datetime and allows data to be None when a Pub/Sub message is published without a payload. It also updates the message handler to default data to None and adds comprehensive unit tests to verify these behaviors. I have no feedback to provide as there are no review comments.

[CorieW]

LGTM

nits:

• data="" still makes .json raise, while data=None returns None. Should "" return None too, in case the emulator sends attribute-only messages that way?
• data: str | None may cause new mypy/pyright warnings for users (e.g. b64decode(msg.data)). Worth a changelog note?
• No test covers a present orderingKey (e.g. "abc" → ordering_key == "abc").
• The older tests could reuse _raw_event instead of building the CloudEvent by hand.

[ajperel]

Agree with Corie on "data: str | None may cause new mypy/pyright warnings for users (e.g. b64decode(msg.data)). Worth a changelog note?"

In the extreme should we consider it a breaking change with the type widening and say change(pubsub) ?

[ajperel]

AI is suggesting that we instead do

raw_data = message_dict.get("data")
message_dict["data"] = raw_data if raw_data else None

Emulator vs. Production Payload Normalization ("" vs. None)

Context: Production GCP Pub/Sub omits the "data" field from CloudEvents for attribute-only messages (resulting in message.data is None), whereas the Firebase Local Emulator emits {"data": ""} (resulting in message.data == "").
Observation: While Message.json now returns None for both cases, user code inspecting message.data directly will observe "" in local emulator runs and None in production GCP. Under Python, b64decode("") succeeds locally (b""), but b64decode(None) crashes in production with TypeError.

This also satisfies the exact docstring contract: "or None if the message was published without a payload".