R devel - Sep 2015 - Long vectors: Missing values and R_xlen_t?

Is there a missing value constant defined for R_xlen_t, cf. NA_INTEGER
(== R_NaInt == INT_MIN) for int(eger)?  If not, is it correct to
assume that missing values should be taken care/tested for before
coercing from int or double?

Thank you,

Henrik

On Sep 20, 2015, at 3:06 PM, Henrik Bengtsson <henrik.bengtsson at
ucsf.edu> wrote:
> Is there a missing value constant defined for R_xlen_t, cf. NA_INTEGER
> (== R_NaInt == INT_MIN) for int(eger)?  If not, is it correct to
> assume that missing values should be taken care/tested for before
> coercing from int or double?
> 
R_xlen_t is type of the vector length (see XLENGTH()) and as such never holds a
missing value (since there is no such thing as a missing length). It is *not* a
native type for R vectors and therefore there is no official representation of
NAs in R_xlen_t.

Although native R vectors can be used as indices, the way it typically works is
that the code first checks for NAs in the R vector and only then converts to
R_xlen_t, so the NA value is never stored in R_xlen_t even for indexing.

--- cut here, content below is less relevant ---

That said, when converting packages from "legacy" .Call code before
long vector support which used asInteger() to convert an index I tend to use
this utility for convenience:

static R_INLINE R_xlen_t asLength(SEXP x, R_xlen_t NA) {
    double d;
    if (TYPEOF(x) == INTSXP && LENGTH(x) > 0) {
        int res = INTEGER(x)[0];
        return (res == NA_INTEGER) ? NA : ((R_xlen_t) res);
    }
    d = asReal(x);
    return (R_finite(d)) ? ((R_xlen_t) d) : NA;
}

Note that this explicitly allows the caller to specify NA representation since
it depends on the use - often it's simply 0, other times -1 will do since
typically anything negative is equally bad. As noted above, this is not what R
itself does, so it's more of a convenience to simplify conversion of legacy
code.

Cheers,
Simon

On Mon, Sep 21, 2015 at 11:20 AM, Simon Urbanek
<simon.urbanek at r-project.org> wrote:
>
> On Sep 20, 2015, at 3:06 PM, Henrik Bengtsson <henrik.bengtsson at
ucsf.edu> wrote:
>
>> Is there a missing value constant defined for R_xlen_t, cf. NA_INTEGER
>> (== R_NaInt == INT_MIN) for int(eger)?  If not, is it correct to
>> assume that missing values should be taken care/tested for before
>> coercing from int or double?
>>
>
> R_xlen_t is type of the vector length (see XLENGTH()) and as such never
holds a missing value (since there is no such thing as a missing length). It is
*not* a native type for R vectors and therefore there is no official
representation of NAs in R_xlen_t.
>
> Although native R vectors can be used as indices, the way it typically
works is that the code first checks for NAs in the R vector and only then
converts to R_xlen_t, so the NA value is never stored in R_xlen_t even for
indexing.
>
> --- cut here, content below is less relevant ---
>
> That said, when converting packages from "legacy" .Call code
before long vector support which used asInteger() to convert an index I tend to
use this utility for convenience:
>
> static R_INLINE R_xlen_t asLength(SEXP x, R_xlen_t NA) {
>     double d;
>     if (TYPEOF(x) == INTSXP && LENGTH(x) > 0) {
>         int res = INTEGER(x)[0];
>         return (res == NA_INTEGER) ? NA : ((R_xlen_t) res);
>     }
>     d = asReal(x);
>     return (R_finite(d)) ? ((R_xlen_t) d) : NA;
> }
>
> Note that this explicitly allows the caller to specify NA representation
since it depends on the use - often it's simply 0, other times -1 will do
since typically anything negative is equally bad. As noted above, this is not
what R itself does, so it's more of a convenience to simplify conversion of
legacy code.
Thank you Simon,

all this helped clarify it for me.  It's in line with what I
suspected, but it is really useful to hear it from the "officials".

Cheers,

Henrik
>
> Cheers,
> Simon
>