[barlowa124]

Fixes #1499.

In MoranProcess with mode="db", _matchup_indices() built matchups
from two different index spaces: enumerate(sources) produced positions
in the dead node's neighbour list, while self.index[target] produced
global player positions. score_all() then played and credited the wrong
players. Nothing raised, which kept it hidden.

Birth-death mode was unaffected only by accident: there sources covers
every vertex, so the enumerate counter coincides with self.index[source].

The change uses self.index[source] in both modes, so matchup pairs are
always global player positions. On the issue's reproduction (5-cycle,
player 2 dead) matchups go from {(0, 0), (1, 4)} to {(1, 0), (3, 4)}:
the dead node's neighbours (players 1 and 3) each play their own
out-neighbours, and the dead vertex is skipped.

A regression test pins the matchup set and the resulting scores in the
exact internal state __next__ creates before score_all().

pytest axelrod/tests/unit/test_moran.py passes: 44 tests.

[marcharper]

The test is failing because the black formatter needs to be run.

[Evreu1pro]

Took this for a spin locally on top of 153335c: test_moran.py passes (44 passed) and the repro from #1499 now gives {(1, 0), (3, 4)} / [0.0, 5.0, 0, 0.0, 5.0], as expected. The black failure is just the players list in the new test. python -m black -l 80 axelrod/tests/unit/test_moran.py (the same -l 80 tox uses) splits it one player per line with a trailing comma, and moran.py itself is already clean. Not 100% sure the cancelled jobs are only fallout from that one failure, but that's the only thing black flags for me.

[drvinceknight]

Thanks for this @barlowa124. The only thing stopping CI from passing is the formatting of the new test. Could you run:

python -m black -l 80 axelrod/tests/unit/test_moran.py

and push the result?

[barlowa124]

Done in 73fb176. Ran python -m black -l 80 axelrod/tests/unit/test_moran.py and pushed. Watching CI.