Skip to content

problem typcasting return of R_ExternalPtrAddr(SEXP s)

2 messages · wsetzer@mindspring.com, Brian Ripley

#
In the odesolve routine lsoda(), I allow the function (named func) that calculates the system of differential equations to be defined in a dll that has been dynamically loaded from the file named in dllname.  I use getNativeSymbolInfo(func, dllname)$address to get the address of the function  and pass it to a C function called via the .Call interface.  Inside that C function,  I use R_ExternalPtrAddr(deriv_func) to get the function pointer.  This requires typcasting the return value of R_ExternalPtrAddr from (void *) to (deriv_func *), defined in a typedef in call_lsoda().  

To be more explicit (I'm replacing irrelevant arguments with ellipses):
In lsoda(...,func,dllname,...), func contains a string giving the name of a function which is defined in the dll whose name is contained in dllname.

inside lsoda() are lines like
func <- getNativeSymbolInfo(symbol.C(func),PACKAGE=dllname)

.Call("call_lsoda",...,func)
 
inside call_lsoda.c are the lines:
SEXP func;

typedef void deriv_func(long int *, double *,double *, double *);

derivs = (deriv_func *) R_ExternalPtrAddr(func);


This all worked up to now, but now, (as Kurt Hornik kindly informs me) gcc 3.4 gives the warning:

warning: ISO C forbids conversion of object pointer to function pointer type

so I need to clean this up so odesolve will compile cleanly for version 2.0.0 of R.  I'm at a loss of what to do, though, since the return value of R_ExternalPtrAddr is (void *), and I need a function pointer.

I'd guess there would need to be a special function (say declared as 
( R_ExternalFnPtr(SEXP s) *)(), 
or something like that), to make all this work, but I can't find such a function.   My C programming is pretty rusty, so I'm wondering if someone has an idea about how I can get around this problem (or perhaps I've misunderstood the meaning of the error message? )

Woody Setzer
#
It's a warning, and occurs several times in R itself.  I don't think it is 
fatal to live with it.

There exists systems on which function pointers are not simple pointers, 
but probably none of them run R -- and they do not support dynamic loading 
of packages' DLLs if they do.
On Mon, 20 Sep 2004 wsetzer@mindspring.com wrote: