[Rd] (PR#9964) package Geneland / Rgui under windows

[Prof Brian Ripley]

I can confirm that if the Geneland is built under gcc 3.4.5 it works in R 
2.6.0 in Rgui (modulo the output which is sent nowhere).  So the issue is 
the Fortran code of the package, and very likely its use of Fortran I/O.

Since this confirms there is not a bug in R 2.6.0, I am filing this on 
R-bugs and closing the bug report.  There is a bug in the package, but 
that is not a subject for R-bugs.

I've put my build up on the CRANextras repository: Uwe might like to 
remove his so that users get that one until such a time as the I/O issues 
are resolved.

On Tue, 16 Oct 2007, Prof Brian Ripley wrote:

> On Mon, 15 Oct 2007, Sean Davis wrote:
>
>> Gilles GUILLOT wrote:
>>> Hi,
>>> I experienced a problem with the package Geneland under R 2.6.0
>>> with  windows XP professional.
>>>
>>> The commands below should simulate a dataset,
>>> then make an MCMC simulation stored in tempdir().
>>>
>>> It works with R 2.5.1 (both GUI and command line)
>>> It works with the command line of R 2.6.0
>>> but not with the R GUI of 2.6.0: no output file is created in tempdir()
>>> and R remains frozen.
>>> I reported it as a bug
>>> (PR#9964)  but did not get any feed back.
>>
>> I think the general rule-of-thumb is to contact the package author for
>> problems with individual packages.  Many package authors read this list,
>> but there are probably some that do not.  Sometimes problems with
>> individual packages are actually bugs in R, but I would say that this is
>> not usually the case.
>
> Overwhelmingly so: I don't believe we know of any in the transition to
> 2.6.0 that were not bugs in the packages concerned.
>
>> However, the package author is probably the best person to make this
>> judgment.
>
> In this case 'Gilles GUILLOT' completely failed to say that he is the
> maintainer, including in his bug report (and he is using a different email
> address from the one in the package: I only know this is the same person
> from a private reply).
>
> I think he needs to run R under a debugger and find out what 'freezes'
> means.  If as I suspect this is in the compiled code of the package, it
> is not a bug in R.
>
> Windows binaries for R 2.6.0 are built with a different compiler than
> 3.4.5, and this has triggered a few package problems (note, not problems
> in R but in the packages concerned):
>
> - a couple of packages infinite loop because the calculations are done in
> extended precision registers and so do not terminate, something we have
> had to work around for LAPACK, for example.
>
> - I think three packages infinite-loop because they write outside array
> bounds.  In one case it always did, but used to get away with it.
>
> None of these explain a Rgui/Rterm difference, if that is what is meant
> here (R has a 'command line' in both, and Rterm is a 'console application'
> in Windows-speak).  The main possible difference relates to I/O, and I see
> that Geneland uses Fortran I/O.  Fortran I/O to '*' (or units 5/6) is
> incompatible with a GUI application, and in gcc 4.2.1 Fortran I/O is
> pretty much incompatible with C I/O (initializing has been known to break
> C I/O).  So I suggest that Fortran I/O is removed and replaced by calls to
> realpr etc.
>
> Filing a bug report on issues in your own package is not at all good form:
> R-bugs is not for 'feed back', but there *was* a comment filed on PR#9964.
>
>

-- 
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