[ESS] ESS + R+ gfortran on windows crash on dyn.unload()

[Richard Arnold]

I am running R-2.6.1 under Windows XP SP2 (2002) [Toshiba,  Intel 
Celeron 1.4Ghz, 704Mb RAM] with MSYS+MinGW(Rtools)
using emacs (22.1) with ESS (version 5.3.5) which I invoke with 
inferior-R-args = "--no-save"

I have the following Fortran routine in a file ff77.f

      SUBROUTINE G(X,FVAL)
      REAL*8 X, FVAL
      FVAL = X**2
      WRITE(6,*) 'X,G(X)=',X,FVAL
      CALL FLUSH
      END

and the following R code in a file test.q

g <- function(x) .Fortran("g",as.double(x),as.double(0))[[2]]
system("R CMD SHLIB ff77.f")
dyn.load("ff77.dll")
print(is.loaded("g"))
print(g(4))
dyn.unload("ff77.dll")
print(is.loaded("g"))
print(sqrt(2))  # just to have something to print
cat("Finished\n")

If I run the R code directly from the operating system command line (in 
MSYS) with

R --no-save < test.q

I get the output:

 > g <- function(x) .Fortran("g",as.double(x),as.double(0))[[2]]
 > gfortran-sjlj   -O3  -c ff77.f -o ff77.o
gcc-sjlj  -std=gnu99  -shared -s  -o ff77.dll ff77.def ff77.o 
ff77_res.o  -Lc:/PROGRA~1/R/R-26~1.1/bin   -lgfortran -lR
system("R CMD SHLIB ff77.f")
 > dyn.load("ff77.dll")
 > print(is.loaded("g"))
[1] TRUE
 > print(g(4))
 X,G(X)=   4.00000000000000        16.0000000000000    
[1] 16
 > dyn.unload("ff77.dll")
 > print(is.loaded("g"))
[1] FALSE
 > print(sqrt(2))
[1] 1.414214
 > cat("Finished\n")
Finished
 >

This is fine - everything works as it should
I now type open R in emacs, make a trivial edit to f77.f to force it be 
recompiled,  and put each line of test.q into R one by one using 
eval-line-and-step and this happens

 > g <- function(x) .Fortran("g",as.double(x),as.double(0))[[2]]
 > system("R CMD SHLIB ff77.f")
gfortran-sjlj   -O3  -c ff77.f -o ff77.o
gcc-sjlj  -std=gnu99  -shared -s  -o ff77.dll ff77.def ff77.o 
ff77_res.o  -Lc:/PROGRA~1/R/R-26~1.1/bin   -lgfortran -lR
 > dyn.load("ff77.dll")
 > print(is.loaded("g"))
[1] TRUE
 > print(g(4))
 X,G(X)=   4.00000000000000        16.0000000000000    
[1] 16
 > dyn.unload("ff77.dll")
 >

Process R finished at Fri Nov 30 10:44:26 2007

i.e. the fortran code works but R crashes at dyn.unload() - (and if I 
don't use inferior-R-args= --no-save then I get asked repeatedly and 
indefinitely about saving the workspace image - and have to terminate 
Rterm using the Task Manager)

If instead I open R in emacs and go
     source("test.q")
then this happens:

 > source("test.q")
gfortran-sjlj   -O3  -c ff77.f -o ff77.o
gcc-sjlj  -std=gnu99  -shared -s  -o ff77.dll ff77.def ff77.o 
ff77_res.o  -Lc:/PROGRA~1/R/R-26~1.1/bin   -lgfortran -lR
[1] TRUE
 X,G(X)=   4.00000000000000        16.0000000000000    
[1] 16
[1] FALSE
[1] 1.414214
Finished
 >

Process R finished at Fri Nov 30 10:47:45 2007

i.e. the code runs but R still exits without being asked to do so.

The problem seems be caused by the WRITE(6,*) statement.  If I comment 
it out then all three approaches work with no trouble.

In summary:
+ the fortran code compiles OK, and always runs when R is invoked from 
the operating system
+ it runs but R exits unexpectedly when dyn.unload() is called when 
running R from inside emacs
Also note:
+ calls to WRITE(6,*) of this form worked fine with emacs+ESS under g77 
(as opposed to gfortran used here) [and the CALL FLUSH command wasn't 
needed to flush the buffer and prevent the WRITE output being deferred 
until R exits]
+ I get identical behaviour whenever I try to use the 
ALLOCATE/DEALLOCATE commands in .f95 code: works under the operating 
system, but not under emacs/ESS - so it's not just a problem with WRITE()

I can live without WRITE() [calls to C routines using printf() have no 
trouble - I've tried this], but really want to be able to use ALLOCATE 
in .f95 - but both WRITE and ALLOCATE seem to cause the same problem.

My question is: what is ESS doing differently form the operating system 
- and is there some way to stop it?

Any help appreciated!
Thanks
Richard