[R-pkg-devel] error in sample(), invalid 'size' argument

[Iñaki Ucar]

On Wed, 15 Jan 2020 at 13:58, Roberts, David <droberts using montana.edu> wrote:
>
> Colleagues,
>
>     I have a function (maxsimset) in package optpart that has worked for
> at least a decade, passes check --as-cran on my ubuntu-based system, and
> builds without errors at win-builder and CRAN windows, but fails at CRAN
> debian (unstable) with the error
>
>  > mss.test(mss.10,shoshsite$elevation)
> Error in sample.int(length(x), size, replace, prob) :
>    invalid 'size' argument
> Calls: mss.test -> sample -> sample.int
>
> an excerpt of the function is below
>
>   1  mss.test <- function(mss,env,panel='all',
>   2              main=deparse(substitute(env)),...)
>   3  {
>   4      if (class(mss) != "mss") {

Not related to the issue, but please use "inherits" instead.

>   5          stop("You must pass an object of class mss ...")
>   6      }
>   7      if (!is.numeric(env)) {
>   8          stop("You must pass only numeric vectors ... ")
>   9      }
> 10      setsiz <- ncol(mss$member)
> 11      nset <- mss$numset
> 12      null <- rep(0,nset)
> 13      for (i in 1:nset) {
> 14          tmp <- sample(1:length(env),setsiz)
> 15          nullmin <- min(env[tmp])
> 16          nullmax <- max(env[tmp])
> 17          null[i] <- nullmax - nullmin
> 18      }
> 19         .     .      .
> 20         .     .      .
> 21         .     .      .
> 22  }
>
> The error occurs at line 14 in the excerpt above.  As you can see, the
> 'size' argument is setsiz, taken at line 10 as the ncol(mss$member); mss
> has already been checked for conformance

No, it has not. Nothing stops you from changing the contents of a S3
object, so just verifying the class is not a conformance check.

> and does indeed have a list object matrix called $member.

What do you mean by "list object matrix"?

> In my tests, setsiz has mode numeric,
> class integer, and size 1 as expected.  The specific error message I
> receive can be generated from sample() if size is a list object or
> data.frame, but not a vector or matrix.

My guess is that, in some call, you got an unintended simplification
when assigning $member. For any object without a "dim" attribute,
ncol(object) is NULL. And a NULL size in sample() raises such an
error.

Iñaki