An NA in any argument disables length validation in the skewnorm/skewlnorm d/p/q/r functions

[StefanoMezzini]

Problem

All eight density/probability/quantile functions in R/skewnorm.R and R/skewlnorm.R guard the length check behind an NA check:

if (lengths >= 4) {
  nas <- any(is.na(q), is.na(mean), is.na(sd), is.na(shape))
  if (!nas) chk_compatible_lengths(q, mean, sd, shape)
}

If any argument contains an NA, chk_compatible_lengths() never runs. Incompatible lengths then flow through to mapply/sn::*, which recycle silently (with a base R warning) and produce output of the wrong length with NAs in the wrong positions.

The NA masking that follows compounds it — na_shape is length(shape) but is used to index a result of max(length(...)):

p[na_shape] <- NA_real_

Reproduction

Adding an NA turns a hard error into silently wrong output:

pskewnorm(1:6, 0, 1, c(1, 2, 3, 4))
#> Error: ... objects must be all zero length or the same length with some of
#>   length of 1 but not lengths 1, 4 and 6.

pskewnorm(1:6, 0, 1, c(NA, 1, 2, 3))
#> Warning: longer argument not a multiple of length of shorter
#> [1]        NA 0.9550173 0.9973002 0.9999367        NA 1.0000000

Only element 1 should be NA; element 5 is a recycling artefact. dskewnorm behaves the same way:

dskewnorm(1:6, 0, 1, c(NA, 1, 2, 3))
#> Warning: number of rows of result is not a multiple of vector length (arg 2)
#> [1]           NA 1.055253e-01 8.863697e-03 2.676605e-04           NA 1.215177e-08

Affected

dskewnorm, pskewnorm, qskewnorm, dskewlnorm, pskewlnorm, qskewlnorm (and rskewnorm/rskewlnorm, which use the same if (!nas) pattern around their own length check).

Suggestion

The length check is a statement about argument lengths and doesn't need to care whether the values are NA — chk_compatible_lengths() should be able to run unconditionally. If it was gated for a reason, the NA masks should at least be recycled to the result length first (e.g. rep_len(is.na(shape), n)), the way dskewlnorm already does for x via xr <- rep_len(x, length(log_lik)).