The semantics of Artifact updates are underspecified in the technical documentation

[aickin]

I read through the documentation about artifact updates in the Streaming Support section, and it feels like this section needs to be clearer about how updates happen. Here are some questions I have that I don't think are currently answered:

1. First off, are names for Artifacts unique in the scope of a Task? Are indexes unique? Is it allowed in the spec to have two Artifacts in a Task with the same name but different indexes? Conversely, is it allowed in the spec to have two Artifacts in a Task with different names but the same index?
2. Assuming that index is unique in the scope of a Task, what do you do when a TaskArtifactUpdateEvent sends a new name for the Artifact? Does it overwrite the previous name? Is it an error? Or is it ignored? Similarly, is a new description for the Artifact an overwrite/error/ignored?
3. When you send metadata in a TaskArtifactUpdateEvent, does it replace the existing metadata for the Artifact, or is it merged into the existing metadata for the Artifact on a field-by-field basis, or is it an error, or is it ignored?
4. How do append and lastChunk work in practice? A few examples that I'm unsure about:
   1. Let's say you send one update with append: false and lastChunk: false, and then send a second update again with append: false and lastChunk: false. Intuitively, I would think that this wipes out and overwrites the first update. Is that correct, or is this an error?
   2. Now, let's say you send several updates in a row with append: true and lastChunk: false. Then you send an update with append: true and lastChunk: true. Finally you send an update with append: true and lastChunk:false. Is that an error?
   3. Similar to the last one, let's say you send several updates in a row with append: true and lastChunk: false. Then you send an update with append: true and lastChunk: true. Finally you send an update with append: false and lastChunk:false. Is that an error or does it wipe out all the previous updates?

I know that some of these may seem like nitpicks, but I tend to think a spec should be really clear about error cases and state transitions. I think in it's current state, a reasonable engineer could assume several different ways this artifact update mechanism works.