Quartile computed the median of empty slices for inputs with fewer than
three elements and discarded the resulting error, returning NaN quartiles
with a nil error. The silent NaN then leaked into InterQuartileRange,
Midhinge and Trimean. Propagate the error instead, mirroring the existing
EmptyInputErr handling.
Quartile called Median on empty half-slices when the input held one value
and discarded both errors, so it returned {NaN, v, NaN} with a nil error.
The NaN spread silently through InterQuartileRange, Midhinge and Trimean.
A single observation is its own lower hinge, median and upper hinge, which
is what Percentile already returns for the same data. Return that instead,
and note why the remaining discarded Median errors cannot fire.
dualfroz
upstream