Single Precision #1
Description
Mutation-selection equilibrium (MSE) and site frequency spectrum (SFS) calculations are currently done in single precision. So for SFS, there may be very slight variations in the reported SFS for the same simulation with the same random number seeds.
Similarly, when starting a simulation in MSE, there can be large changes in the simulation result for the same simulation with the same random number seed due to very slight changes in the MSE (the RNG randomly picks out the change and each mutation's path through the simulation after the change is affected).
I am considering several solutions for this, including leaving it alone as it is not strictly a bug - the reported results are still consistent with theory and good (as far as I can tell, however, I will be thoroughly checking). Other solutions are to round the MSE/SFS calculations to 5 significant digits and to switch to double precision for the summation/integration steps.