In space events, is the `via` required or optional?

[bnjbvr]

Link to problem area:

• https://spec.matrix.org/unstable/client-server-api/#mspacechild
• https://spec.matrix.org/unstable/client-server-api/#mspaceparent

Issue
What is wrong? How can we improve?

For both events, it is indicated that the via field is required, and a few lines below, it reads:

When not present or invalid, the room is not considered to be part of the parent space.

The first part of the sentence, namely When not present, makes it sound like it's optional.

Is it that:

• the field is actually optional, so we shouldn't mark it as required?
• or the event is invalid when the field is not present, and thus the room is not considered to be part of the parent space?

[richvdh]

[links to stable spec are recommended, in case the anchors change in future versions]

Is it that:

* the field is actually _optional_, so we shouldn't mark it as _required_?
* or the event is invalid _when_ the field is not present, and thus the room is not considered to be part of the parent space?

Gut feeling is that, from a client's point of view, the two are basically equivalent, so the question seems a bit moot, but let's dig a bit deeper:

I suppose the real question here is: "Suppose I was implementing a server, that did its best to ensure all events sent by clients were 'valid' according to the schema. Should I allow clients to send m.space.child|parent events with no via?"

... and the answer to that question would appear to be "yes", so I suppose the answer to your original question is "via is actually optional".

[clokep]

I think the expectation was that it has via and optionally order/recommended to be a valid link between room and space or have an empty body to cancel the link between room and space.

So from an OpenAPI perspective, it is optional.

[bnjbvr]

So, just to make sure that I understand correctly: it is valid that a client sends or receives such space events, where the via property is missing? i.e., those events shouldn't be considered invalid by clients, as these are legal JSON encodings?

As a result: should #1618 be reverted?

[richvdh]

Yes it is valid that a client sends space events with no via property. Servers should allow them to do so, and #1618 should probably be reverted for that reason.

On the subject of a client receiving such an event: an event with no via property should be treated the same as an invalid event, so the question seems to me to be completely moot.

[bnjbvr]

I don't think the question is moot: some clients might decide to render an "unable to parse such event" item, for an event that was malformed or invalid according to the spec. For clients which don't make a distinction between an invalid space event, and a space event with an empty content (i.e. missing via field), then indeed it doesn't change semantics, but it breaks the "explicit is better than implicit" principle.