Much of the initial code would implicitly round array indices (or similar Int-only values) in from floating point representations.
@ma-laforge was not 100% sure what happened in the original execution, so the rounding was wrapped into a call to
array_round(i) = round(Int, i). It is not certain if this function correctly reproduces the original behaviour.
Either way, these calls should be replaced with explicit rounding functions, for example:
x[1:div(n,2)] instead of x[1:(n/2)] - Well, unless the original code expected rounding up.x[1:round(Int, float_idx)]x[1:floor(Int, float_idx)]x[1:ceil(Int, float_idx)]
Unfortunately, the original author's intent for rounding is not always easy to decipher.
Floating-point order
There were other cases where order was passed around as a floating point value (potentially having a 0.5 fractional value after dividing by 2).
Goal
The original intent should be resolved so that floating point representations are eliminated. As a side benefit, this should also reduce accidental errors introduced when modifying code in the future.
Much of the initial code would implicitly round array indices (or similar
Int-only values) in from floating point representations.@ma-laforge was not 100% sure what happened in the original execution, so the rounding was wrapped into a call to
array_round(i) = round(Int, i). It is not certain if this function correctly reproduces the original behaviour.Either way, these calls should be replaced with explicit rounding functions, for example:
x[1:div(n,2)]instead ofx[1:(n/2)]- Well, unless the original code expected rounding up.x[1:round(Int, float_idx)]x[1:floor(Int, float_idx)]x[1:ceil(Int, float_idx)]Unfortunately, the original author's intent for rounding is not always easy to decipher.
Floating-point
orderThere were other cases where
orderwas passed around as a floating point value (potentially having a 0.5 fractional value after dividing by 2).Goal
The original intent should be resolved so that floating point representations are eliminated. As a side benefit, this should also reduce accidental errors introduced when modifying code in the future.