A0-4559: Create units with ancient parents

[timorleph]

This should finish the, somewhat unaccurately named, "censorship resistance" project.

[github-actions]

Please make sure the following happened

• Appropriate tests created
• Infrastructure updated accordingly
• Updated existing documentation
• New documentation created
• Version bumped if breaking changes

[Marcin-Radecki]

Changes looks good.

There are two disabled tests, I tried to enable them on this branch, they pass but unfortunately I wrote them down in a way they do not test what they should be testing. I'll create ticket for fixing this

[Marcin-Radecki]

The unit test coverage in this module is neat already, however we can consider adding a test to cover this line as well, e.g. add this code to existing following_inherits_units test (at the end):

        units_collector.add_unit(&units[1][3]);
        let parents = units_collector
            .prospective_parents(NodeIndex(0))
            .expect("we should be able to retrieve parents");
        assert_eq!(parents.item_count(), 4);
        let new_units: Vec<_> = units[1]
            .iter()
            .map(|unit| (unit.hash(), unit.round()))
            .collect();
        let selected_parents: Vec<_> = parents.values().cloned().collect();
        assert_eq!(new_units, selected_parents);

[timorleph]

This is already tested by the units produced by 0-2 being inserted already though, no?

[Marcin-Radecki]

Also, this flow is not tested in this module - maybe some simple test which tries to add unit with round == self.for_round, which does not change state of UnitsCollector?

Just realized it is tested in creator module in creates_unit_with_ancient_parents_weird_order when we try to add round 1 unit to UnitCollector which has round_for == 0. So this comment becomes very minor.

[timorleph]

Added an explicit test for that. Also, I think the test in creator doesn't do that, the lowest round_for we can get is 1?

[Marcin-Radecki]

You mean here there is sixth parent? Or there are ancient parents?

[timorleph]

Ah, good point, too much copypasting, fixed.

[Marcin-Radecki]

This is the only piece of the code I thinkied longer if it's required, but I think now it is, we just need to have various scenarios covered in tests. As we do two loops now right, first one iterates by rounds and second one creates UnitCollectors if necessary

1. We have not enough unit collectors, e.g. current_round is 2, but unit.round() is 6

2. We have enough unit collectors, e.g. current_round is 6, but unit.round() is 2.

3. I think is covered already, but 2 is not? E.g. we can have minimal dag up to 6 rounds, and then call this add_unit method for round 2 unit for NodeIndex which was not in the dag yet, and then create round 7 unit and then check parents etc.

[timorleph]

Point 1 cannot really happen, we are missing at most one round, because the units come in order to the creator. I did add one slightly better test anyway, to ensure this loop works properly though.

[timorleph]

Also, activated the two tests you mention.

(Alsoalso I forgot to fmt, but I'm fixing that. :p)