[Rd] properly reporting compile bugs

[Prof Brian Ripley]

You should report package problems to the package maintainer (see the R 
posting guide and FAQ).  *But* it seems you are just replicating what the 
CRAN daily checks on x86_64 Debian Linux do, and I repeat those checks 
most weeks on Fedora 8 (including twice already this week as we test the 
near-final versions of R 2.7.2).  I see few actual compile errors but 
quite a few check errors (e.g. on the examples).

See http://cran.r-project.org/web/checks/check_summary.html .

On Wed, 13 Aug 2008, someone ashamed of her real name wrote:

> A while ago I decided to simply download and install every package on CRAN
> (disk space is cheap these days...).

Yes, quite a few people do that.  My dept has central R package 
installations for both i386 and x86_64 F8 of most of CRAN and BioC 2.2.

> In the process, I kept loose track of which packages generated one or 
> more 'compilation errors' during the install process (machine runs 
> 64-bit Fedora 8, gcc 4.1 etc. etc). Most of the errors seemed to occur 
> in packages I have little need for at present, but as a courtesy, I 
> thought it might be helpful to report these 'errors' - but, how, and to 
> whom (package maintainer)? Is there a preferred (official) mechanism for 
> this? A pointer to the relevant webpage would be more than sufficient.
>
> Some of the errors or lib dependencies, which are perhaps not generally
> useful to report. But some are outright code errors - either syntax
> problems, or deprecated bits of C and Fortran.

That is really rather surprising, and may indicate a problem on your own 
setup.

Using C90 or F77 constructs is not an error: it is all that is required of 
R compilers. Remember that to get on CRAN a package has had to compile on 
at least one platform, and the vast majority have compiled and run their 
tests on at least four (i686 Linux, x86_64 Linux, Windows, Mac OS X 
i386/ppc).  Right now the only non-Windows-only CRAN packages that will 
not install for me on x86_64 F8 Linux are RDieHarder, Rmpi, ROracle, 
RQuantLib, RScaLAPACK, Rcplex and Rlsf.

For Rmpi that is a fairly recent Fedora packaging issue and can be 
resolved by editing the configure script (and this has been reported to 
the package maintainer and discussed on this list).

There are quite a number of packages with known C++ bugs that g++ 4.1.1 
accepts.  We have tested on stricter C++ compilers and reported to the 
maintainers -- some of these also fail with g++ 4.3.x (which I believe the 
CRAN checks now uses).

> Thanks in advance...
>
> 	[[alternative HTML version deleted]]

If you do write to maintainers it is discourteous to send HTML mail (and 
the R posting guide explicitly asked you not to send it here).

We prefer real names and addresses here, so please don't write to 
maintainers without a proper signature block (again, see the R posting 
guide).

-- 
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