Talk:Req:Subsecond

R2

 * 1) Define "coherent model", for record-keeping purposes. --user:mhauer
 * 2) * Definition section added. Coherent means all objects are updated to the same time but not necessarily consistent state.  This points out a lack of clarity in the different requirement for event->delta vs. delta->event. R2 is amended to address this issue. --user:dchassin

R3

 * 1) 2^10 = 1024 ~= 10^3, so 10^9 precision only requires 2^30 -> an int32.  I doubt we need to go down to attoseconds. --user:mhauer
 * 2) * 10^-9 is nanosecond scale, which seems ample for the kind of physics we would model. Note also there is no requirement that 1 second be an upper bound on how long delta mode can run.  So using int64 internally may be needed to longer runs in delta mode if desired. --user:dchassin
 * 3) We may want to include a 'global_subsecond_precision' that explicitly states how many mini-ticks are in a second ... or we can implicitly hardcode nanoseconds and bludgeon the naysayers. --user:mhauer
 * 4) * Using variable precision is the same issue we got into with the original tick implementation. It was ignored by implementations and now can't be used anyway. The spec calls out what the delta timestep is as a global variable. This is distinct from what the storage resolution of subsecond timesteps is, which is pinned down at 10^-9 seconds. --user:dchassin

R5

 * 1) This could be reworded to emphasize that only the classes that support finite-time difference updates shall be required to implement an update function callback. --user:mhauer
 * 2) * Done. --user:dchassin

R6

 * 1) See above on setting a flag if they want it. --user:mhauer
 * 2) * I think this is handled in the spec. --user:dchassin