[R] Make check fails on d-p-q-r-tests.R...

[Prof Brian Ripley]

I am afraid this does look like a real problem, if a minor one. We have 
for the first problem

x <- 10^(ex <- c(1,2,5*(1:5),50,100,200,300,Inf))
ex <- -c(rev(1/x), ex)
qcauchy(ex, log=TRUE)

The first entry of the result should be Inf and the last -Inf.  From the 
output you have shown us I would guess the last is NaN.

If this is what is going on, it is a problem but a somewhat esoteric one 
that probably reflects a lack of IEC60559 (aka IEEE754) conformance.  It 
needs more hands-on debugging than we can provide on a help list.


On Tue, 27 Sep 2005, Jeff Ross wrote:

> On 1:03:39 am 09/27/05 Prof Brian Ripley <ripley at stats.ox.ac.uk> wrote:
>
>> You will have to show us the error!   It will be shown in
>> d-p-q-r-tests.Rout.fail (unless this was a segfault or similar).
>>
>> It is not OK to skip the test, but note that this test is random and
>> does fail about 1 in 50 times, so you could just try rerunning it.
>
> I am so sorry.  I didn't even think to look for a .fail file.
>
> Here is the last bit of it:
>
>>> ## for PR#7902:
>> ex <- -c(rev(1/x), ex)
>> All.eq(-x, qcauchy(pcauchy(-x)))
> [1] TRUE
>> All.eq(+x, qcauchy(pcauchy(+x, log=TRUE), log=TRUE))
> [1] TRUE
>> All.eq(1/x, pcauchy(qcauchy(1/x)))
> [1] TRUE
>> All.eq(ex,  pcauchy(qcauchy(ex, log=TRUE), log=TRUE))
> [1] "`is.NA' value mismatches: 1 in current, 0  in target"
> Warning message:
> NaNs produced in: qcauchy(p, location, scale, lower.tail, log.p)
>> II <- c(-Inf,Inf)
>> stopifnot(pcauchy(II) == 0:1, ## qcauchy(0:1) == II,
> +           pcauchy(II, log=TRUE) == c(-Inf,0),
> +           qcauchy(c(-Inf,0), log=TRUE) == II)
> Error in if (!(is.logical(r <- eval(ll[[i]])) && all(r)))
> stop(paste(deparse(mc[[i +  :
>  missing value where TRUE/FALSE needed
> In addition: Warning message:
> NaNs produced in: qcauchy(p, location, scale, lower.tail, log.p)
> Execution halted
>
> This is from the latest R-patched source tar ball, and it is the identical
> error as R-2.1.1.  I've consistently gotten the same error, even with
> configuration file tweaks and making clean between runs.
>
> Thanks!
>
> Jeff Ross

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595