On Thu, Nov 18, 2021 at 05:49:13PM +0100, Jan Kara wrote: > The first part of the series improves some tests so that they accept > other valid decisions for bisection points. This is needed because to > make it easier to share some logic between normal and stochastic > bisection, I needed to slightly change some bits for normal bisection > and then since commit weights will be computed in a somewhat different > order, also chosen bisection points are sometimes different. I have only looked through a couple of the first half of your patches, but I'm not sure I understand why non-stochastic bisection needs to change at all in order to support stochastic bisection. In other words, if we're tweaking all of these tests to allow picking equivalent bisection points, why can't we simply leave them alone? It would be nice if normal bisection didn't change as a result of adding a new feature on top. Thanks, Taylor
