flow_nums: standard interface for loading/dumping

[oliver-sanders]

The flow_nums define the flows to which a task belongs.

Context:

• Under normal conditions flow_nums = {1} # task belongs to the original flow.
• Reflow tasks differ flow_nums = {2} # task belongs to a new flow.
• When flows merge their numbers merge flow_nums = {1, 2} # flows 1 & 2 merged.
• Triggered tasks are special cases flow_nums = {}.

Because flow_nums are a set order is not guaranteed. Consequently we should take care to serialise / deserialise flow_nums via a standard interface to ensure they always match.

Suggest the following:

def serialise(flow_nums):
   return json.dumps(sorted(flow_nums))

def unserialise(flow_num_str):
    set(json.loads(flow_num_str))

def db_serialise(flow_nums):
    return sorted(flow_nums)

If we don't sort the nums there is a risk that they could come out in a different order than they went in e.g. {2, 1} rather than {1, 2} which would break some internal comparison logic as well as making life harder for the UIs which would prefer one canonical form for each set.

Pull requests welcome!

[hjoliver]

This won't hurt, but I'm not sure it's necessary - any comparison logic ought to take account of the fact that it's comparing (unordered) sets, no?

[oliver-sanders]

This won't hurt, but I'm not sure it's necessary - any comparison logic ought to take account of the fact that it's comparing (unordered) sets, no?

If they are in set form, yes, however, if they are serialised, no, e.g:

#4581 (comment)

It's the read/write to the DB and data-store where this matters not inside the task pool (where set comparisons work great).

[hjoliver]

Ok, duh. Got it 👍🏻