[R] Reloading a shared library with dyn.load

[Duncan Temple Lang]

Prof Brian Ripley wrote:
> As there is AFAIK no way to unregister entry points, you cannot safely
> unload a DLL with registered entry points (which is what your code appears
> to do).  So if you want to do this, don't register them,

This is not true.  The unloading of a shared library automatically
unregisters the entry points (the routine Rf_freeDllInfo in
Rdynload.c). So using the registration is not a problem.

Checking this on both my Linux and Windows boxes running 1.7 patched
doesn't cause any problems.  Adding a static variable to the
HelloFromC routine and both incrementing it and adding its value to
the returned string in each call illustrates that when a second
dyn.load() call is made, a new version of the library comes in.

So I can't reproduce the problem directly from the commands.
Something that comes to mind is that another process has got a lock on
the file. But I don't see that as likely.


> 
> Package tcltk does unload its DLL (so the Tcl/Tk dlls get released) if 
> detached, and so was not modified to register.
> 
> Until un-registration is provided, I would not use the registration 
> mechanisms during development.

This is *NOT* a necessary precaution.  Registration has the potential
to improve this situation rather than diminishing it. So please don't
let this discourage people from using registration.

 D.


> 
> On Fri, 16 May 2003, James Wettenhall wrote:
> 
> > Hi,
> > 
> > I'm using dyn.load to load a shared library (compiled from C 
> > code) into R.  If I dyn.unload it and then dyn.load it again, I 
> > get an hourglass icon in Rgui (R 1.7.0, Win 2000), and it 
> > just sits there forever.  I can't press Escape to stop the 
> > current computation, but I can close Rgui without resorting to 
> > using the Task Manager.
> > 
> > Is it a problem with my use of R_alloc?  Do I need something 
> > like R_cleanup?  Or is it a problem with dyn.load?
> > 
> > I'll demonstrate how I compile a C file, HelloFromC.C into a 
> > shared library, HelloFromC.dll and then load it into R.
> > 
> > HelloFromC.c
> > ------------
> > #include <R.h>
> > #include <Rinternals.h>
> > #include <R_ext/Rdynload.h>
> > #include <R_ext/Memory.h>
> > #include <R_ext/Applic.h>
> > #include <stdio.h>
> > 
> > void HelloFromC(char **result)
> > {
> >    *result = (char *) R_alloc(20,sizeof(char));
> >    sprintf(*result,"Hello from C!"); 
> > }
> > 
> > static const
> > R_CMethodDef CEntries[] = {
> > {"HelloFromC",(DL_FUNC) &HelloFromC,1},
> > {NULL,NULL,0}
> > };
> > 
> > void R_init_HelloFromC(DllInfo *info)
> > {
> >     R_registerRoutines(info,CEntries,NULL,NULL,NULL);
> > }
> > 
> > ----------------------------------------------------------------
> > c:\james\HelloFromC> Rcmd SHLIB HelloFromC
> > making HelloFromC.d from HelloFromC.c
> > gcc   -IC:/JAMES/rw1070/src/include -Wall -O2   -c HelloFromC.c 
> > -o HelloFromC.o
> > ar cr HelloFromC.a *.o
> > ranlib HelloFromC.a
> > gcc  --shared -s  -o HelloFromC.dll HelloFromC.def HelloFromC.a  
> > -LC:/JAMES/rw1070/src/gnuwin32  -lg2c -lR
> > -----------------------------------------------------------------
> > 
> > Now in R 1.7.0...
> > -----------------
> > 
> > > dyn.load("HelloFromC")
> > > result <- ""
> > > .C("HelloFromC",result)[[1]]
> > [1] "Hello from C!"
> > > dyn.unload("HelloFromC")
> > > dyn.load("HelloFromC")
> > 
> > This is where it freezes.
> > 
> > Any suggestions?
> > 
> > Regards,
> > James
> > 
> > ______________________________________________
> > R-help at stat.math.ethz.ch mailing list
> > https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> > 
> 
> -- 
> 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
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help

-- 
_______________________________________________________________

Duncan Temple Lang                duncan at research.bell-labs.com
Bell Labs, Lucent Technologies    office: (908)582-3217
700 Mountain Avenue, Room 2C-259  fax:    (908)582-3340
Murray Hill, NJ  07974-2070       
         http://cm.bell-labs.com/stat/duncan