In the spec, we remember and return the same promises for efficiency if it's semantically correct. The algorithms we have in the spec are representing this. This may look like a requirement to the spec readers. And, we have some unit test items where we actually check this kind of equality.

Do you think this really should be a requirement?

I'm asking this since:

• When implementing the Streams in Blink, we encountered some difficulties with ensuring some of the equality we can infer from the algorithm of the methods/getters.
• There seems to be no real practical needs for such a requirement.