[R] Problem building a DLL from Fortran 90 source under Windows (with solution)

[Prof Brian Ripley]

On Mon, 9 Oct 2006, Mstislav Elagin wrote:

> Hi, All,
>
> I have come across a problem building a DLL from .f90 source (R 2.3.1,
> Windows XP). When using the R CMD SHLIB procedure, the DLL itself was
> being built, but its export table was empty.
>
> Among the output from R CMD SHLIB the following message appeared:
>
> c:\mingw\bin\nm.exe: 'a.out': No such file
>
> The reason is the empty list of dependencies in the pattern rule for
> DLLs contained in the file MkRules:
>
> %.dll:
> 	@$(ECHO) EXPORTS > $*.def
> 	@$(NM) $^ | $(SED) -n 's/^........ [BCDRT] _/ /p' >> $*.def
>
> The dependencies list is constructed in the MakeDll file that contains
> the following:
>
> CFSOURCES=$(wildcard -f *.c *.f *.f90 *.f95)
> CSOURCES=$(wildcard -f *.c)
> CXXSOURCES=$(wildcard -f *.cc *.cpp *.C)
> FSOURCES=$(wildcard -f *.f)
> FCSOURCES=$(wildcard -f *.f90 *.f95)
> OBJSA=$(foreach i,$(CSOURCES) $(FSOURCES) $(CXXSOURCES),$(basename $i).o)
>
> Observe that the neither CFSOURCES nor FCSOURCES in included in OBJSA,
> i.e. an object file made from a f90 source never becomes part of
> dependencies. If I add $(FCSOURCES) to OBJSA, the DLL gets built
> correctly.
>
> I wonder whether there is a reason for not including .f90 (and .f95)
> sources in the list of dependencies or this can be considered a bug in
> R.

Please use a current version of R (as the posting guide asks). From the 
CHANGES file for 2.3.1 patched:

   Objects from Fortran 9X sources no longer need to be declared to R CMD
   SHLIB/INSTALL via OBJS in Makevars.win.


-- 
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