R devel - Mar 2005 - forcing R CMD COMPILE

Hi,

I am developing a package with a lot of C++ code, and I have a question
about R CMD COMPILE. As I can see, when the package's Makefile calls

R CMD COMPILE foo.cpp

R verifies if foo.o exists, and if it is up to date (last modified after
foo.cpp's last), trying to avoid recompile foo.cpp. But __all__ my files
have other dependencies (at least a .h). The Makefile verifies it, and
calls COMPILE only when it is sure that foo.o is not up to dated. The
problem occours, for example, if I modify only foo.h, and then compile
again. R will say that foo.o is up to date, but it is not true.

Is there any way to force R CMD COMPILE?? Well, I can remove the .o file
before COMPILE, but I think R could do something about it...

Thanks,

Pedro Andrade

--
LEG - Laboratório de Estatística e Geoinformação
          http://www.est.ufpr.br/leg/
          Departamento de Estatística
        Universidade Federal do Paraná

On Wed, 23 Mar 2005, Pedro Ribeiro de Andrade Neto wrote:
> I am developing a package with a lot of C++ code, and I have a question
> about R CMD COMPILE. As I can see, when the package's Makefile calls
>
> R CMD COMPILE foo.cpp
Why does your package have a Makefile?  And why is it calling COMPILE and 
not SHLIB?
> R verifies if foo.o exists, and if it is up to date (last modified after
> foo.cpp's last), trying to avoid recompile foo.cpp.
Not quite: COMPILE is a make facility, and it calls make with its own 
Makefiles.  This does not seem appropriate to your usage.
> But __all__ my files have other dependencies (at least a .h). The 
> Makefile verifies it, and calls COMPILE only when it is sure that foo.o 
> is not up to dated. The problem occours, for example, if I modify only 
> foo.h, and then compile again. R will say that foo.o is up to date, but 
> it is not true.
>
> Is there any way to force R CMD COMPILE?? Well, I can remove the .o file
> before COMPILE, but I think R could do something about it...
Better not to use R CMD COMPILE.  R itself does not (AFAICS): it is there 
for S-PLUS compatibility.  It seems only ROracle on CRAN does.

Note that R CMD INSTALL is usually called on fresh sources (an unpacked 
tarball) so dependencies are not usually an issue.


-- 
Brian D. Ripley,                  ripley@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