set_input for one of a variable's own periods (one month of a monthly variable, one year of a yearly one) replaces the value stored for that period, but leaves in place what the simulation had calculated from it for the same variable at periods of another size: the sum calculate caches for a monthly flow over a year, and the twelfth it caches for a yearly flow at a month. A later calculate returns the cached value, so the result depends on whether the longer (or shorter) period was read before the input was set.
Executed on master b78b0ba (two people, formula-less float variables, auto_carry_over_input_variables off):
| Steps |
Result |
calculate("flow_m", "2013"); set_input("flow_m", "2013-05", [50, 0]); calculate("flow_m", "2013") |
[0, 0] |
set_input("flow_m", "2013-05", [50, 0]); calculate("flow_m", "2013") |
[50, 0] |
calculate("flow_y", "2013-05"); set_input("flow_y", "2013", [1200, 0]); calculate("flow_y", "2013-05") |
[0, 0] |
set_input("flow_y", "2013", [1200, 0]); calculate("flow_y", "2013-05") |
[100, 0] |
flow_m is a monthly float variable and flow_y a yearly one, both with no formula.
Where it comes from: Simulation.calculate_add stores the sum at the requested period and calculate_divide stores the twelfth at the month (holder.put_in_cache(result, period, self.branch_name)), and _calculate returns a stored value for the requested period before anything else. Holder.set_input with a period of the variable's own unit calls Holder._set for that period only.
The inputs set over a longer period through set_input_divide_by_period and set_input_dispatch_by_period have the same problem and more; the PR for those helpers drops the variable's calculated sums and twelfths that overlap the input period. This issue is the remaining path: an input for the variable's own period. The same rule would fix it (an input drops the values calculated for that variable at overlapping periods of another size, on the branch it is stored under and the branches read through it), applied in Holder.set_input.
Not covered by #571 (which keeps the cache for a flow read over a period of another unit), #566/#576 (the fast cache, not holder storage) or #560 (other variables calculated from the input, on branches).
set_inputfor one of a variable's own periods (one month of a monthly variable, one year of a yearly one) replaces the value stored for that period, but leaves in place what the simulation had calculated from it for the same variable at periods of another size: the sumcalculatecaches for a monthly flow over a year, and the twelfth it caches for a yearly flow at a month. A latercalculatereturns the cached value, so the result depends on whether the longer (or shorter) period was read before the input was set.Executed on master b78b0ba (two people, formula-less float variables,
auto_carry_over_input_variablesoff):calculate("flow_m", "2013");set_input("flow_m", "2013-05", [50, 0]);calculate("flow_m", "2013")[0, 0]set_input("flow_m", "2013-05", [50, 0]);calculate("flow_m", "2013")[50, 0]calculate("flow_y", "2013-05");set_input("flow_y", "2013", [1200, 0]);calculate("flow_y", "2013-05")[0, 0]set_input("flow_y", "2013", [1200, 0]);calculate("flow_y", "2013-05")[100, 0]flow_mis a monthly float variable andflow_ya yearly one, both with no formula.Where it comes from:
Simulation.calculate_addstores the sum at the requested period andcalculate_dividestores the twelfth at the month (holder.put_in_cache(result, period, self.branch_name)), and_calculatereturns a stored value for the requested period before anything else.Holder.set_inputwith a period of the variable's own unit callsHolder._setfor that period only.The inputs set over a longer period through
set_input_divide_by_periodandset_input_dispatch_by_periodhave the same problem and more; the PR for those helpers drops the variable's calculated sums and twelfths that overlap the input period. This issue is the remaining path: an input for the variable's own period. The same rule would fix it (an input drops the values calculated for that variable at overlapping periods of another size, on the branch it is stored under and the branches read through it), applied inHolder.set_input.Not covered by #571 (which keeps the cache for a flow read over a period of another unit), #566/#576 (the fast cache, not holder storage) or #560 (other variables calculated from the input, on branches).