Specify the amount and multiplicity of the ports for each operations.

[aborgna-q]

With multiports we can now define the exact number of ports each operation will have. There's a number of issues that are not currently specified in the spec:

• We should clarify the order of dataflow/constant/other edges (Clarify port numbering vs signature/consts/others #23)
• We should specify which ports must have exactly one connection vs which can have any number of edges (including none).

In a related note, having a fixed number of ports means that the port counts in the node serialization are redundant. Should we drop them from the serialization specification?

[cqc-alec]

I propose that we keep ports as simple as possible at the spec level, and keep multiports as an implementation detail:

• Edges in the hugr may either have no ports associated with them, or exactly one port at each end.
• Ports always exist on Value and Static edges; never on Hierarchy, Order or Control Flow edges.
• Every port has an associated SimpleType. Joined ports have the same SimpleType.
• A ClassicType outgoing port may have any number of edges associated with it (including zero). A LinearType port, and an incoming port of any type, must have exactly one edge associated with it.

[ss2165]

The spec in my head is aligned with this, what is misaligned? Maybe everything here is not stated explicitly?

[cqc-alec]

Yeah I agree it is aligned, though maybe not fully explicit. This was in response to:

We should specify which ports must have exactly one connection vs which can have any number of edges (including none).

[ss2165]

Yes I think having a section early on (in the Edges section) that enumerates the points you list above would be good, even if strictly redundant.

[ss2165]

I'm in favour of dropping port counts in serialization, as long as we are happy with the constraint:
"All op implmentations, current and future, must be able to report the number of ports they require (for edges of every kind except Hierarchy)."

As far as I can tell that constraint is not too onerous?