When splitting self tasks, consider whether we have stars, not whether we have feedback.

This fixes a rare issue when running with many many more stars than gas.

I feel it is safe but the splitting code is dangerous so an other pair of eyes is welcomed.

So looks OK as we seem to check for stars + count for self tasks and feedback + stars count for pair tasks initially and this follows that logic. Why isn't it stars + count for both task types?

We only have pair tasks when running with feedback. If we have stars in the simulation bu they don't do any feedback on their gas neighbours then we only have self.

Not the clearest distinction ever I agree...

Hmm, so in the runner main I see runner_doself_branch_stars_feedback and runner_dopair_branch_stars_feedback, so I guess the self version is never called.

Anyway, we can accept I think.

Sorry, I wasn't clear. with_feedback always implies with_stars.

If we don't have feedback, we will still need to have some stars-related self-like tasks in the tree to set the super level. Generally speaking, running with stars but without feedback is not common.

Happy if you are.

merged

mentioned in commit 2a0b62ba