R devel - Aug 2019 - Rf_defineVar(symbol, R_UnboundValue, environment) questions

While poking around the C++ code in the dplyr package I ran across the idiom
   Rf_defineVar(symbol, R_UnboundValue, environment)
to [sort of] remove 'symbol' from 'environment'

Using it makes the R-level functions objects(), exists(), and get()
somewhat inconsistent and I was wondering if that was intended.  E.g.,  use
SHLIB to make something from the following C code that dyn.load can load
into R

% cat defineVarAsUnboundValue.c
#include <R.h>
#include <Rinternals.h>

SEXP defineVarAsUnboundValue(SEXP name, SEXP envir)
{
    Rf_defineVar(name, R_UnboundValue, envir);
    return R_NilValue;
}
erratic:bill:292% R-3.6.1 CMD SHLIB defineVarAsUnboundValue.c
gcc -std=gnu99 -I"/home/R/R-3.6.1/lib64/R/include" -DNDEBUG
-I/usr/local/include  -fpic  -g -O2  -c defineVarAsUnboundValue.c -o
defineVarAsUnboundValue.o
gcc -std=gnu99 -shared -L/home/R/R-3.6.1/lib64/R/lib -L/usr/local/lib64 -o
defineVarAsUnboundValue.so defineVarAsUnboundValue.o
-L/home/R/R-3.6.1/lib64/R/lib -lR
erratic:bill:293% R-3.6.1 --quiet --vanilla
> dyn.load("defineVarAsUnboundValue.so")
> envir <- list2env(list(One=1, Two=2))
> objects(envir)
[1] "One" "Two"
>
> .Call("defineVarAsUnboundValue", quote(Two), envir)
NULL
> objects(envir)
[1] "One"
> objects(envir, all.names=TRUE) # is "Two" a 'hidden'
object?
[1] "One" "Two"
> exists("Two", envir=envir, inherits=FALSE)
[1] TRUE
> get("Two", envir=envir, inherits=FALSE) # get fails when exists
says ok
Error in get("Two", envir = envir, inherits = FALSE) :
  object 'Two' not found

Should Rf_defineVar(sym, R_UnboundValue, envir) remove sym from envir?

Bill Dunlap
TIBCO Software
wdunlap tibco.com

	[[alternative HTML version deleted]]

Thanks for spotting this pattern. Using Rf_defineVar() this way to 
delete a variable from package code is wrong because WRE does not 
describe such behavior. That behavior has not been intended by the 
implementation and as you observe, the variable will not be properly 
deleted.

I found another pattern in another package (Rcpp, Rcpp11) which calls 
directly the .Internal() function that deletes a variable, that is also 
wrong, one should only use exported wrappers. The only legitimate way to 
delete a variable in already released versions of R from C is to use 
eval() to call R's rm()/remove(), but we will consider adding a C 
function to the API to do this directly.

Best
Tomas

On 8/15/19 7:12 PM, William Dunlap via R-devel wrote:
> While poking around the C++ code in the dplyr package I ran across the
idiom
>     Rf_defineVar(symbol, R_UnboundValue, environment)
> to [sort of] remove 'symbol' from 'environment'
>
> Using it makes the R-level functions objects(), exists(), and get()
> somewhat inconsistent and I was wondering if that was intended.  E.g.,  use
> SHLIB to make something from the following C code that dyn.load can load
> into R
>
> % cat defineVarAsUnboundValue.c
> #include <R.h>
> #include <Rinternals.h>
>
> SEXP defineVarAsUnboundValue(SEXP name, SEXP envir)
> {
>      Rf_defineVar(name, R_UnboundValue, envir);
>      return R_NilValue;
> }
> erratic:bill:292% R-3.6.1 CMD SHLIB defineVarAsUnboundValue.c
> gcc -std=gnu99 -I"/home/R/R-3.6.1/lib64/R/include" -DNDEBUG
> -I/usr/local/include  -fpic  -g -O2  -c defineVarAsUnboundValue.c -o
> defineVarAsUnboundValue.o
> gcc -std=gnu99 -shared -L/home/R/R-3.6.1/lib64/R/lib -L/usr/local/lib64 -o
> defineVarAsUnboundValue.so defineVarAsUnboundValue.o
> -L/home/R/R-3.6.1/lib64/R/lib -lR
> erratic:bill:293% R-3.6.1 --quiet --vanilla
>> dyn.load("defineVarAsUnboundValue.so")
>> envir <- list2env(list(One=1, Two=2))
>> objects(envir)
> [1] "One" "Two"
>> .Call("defineVarAsUnboundValue", quote(Two), envir)
> NULL
>> objects(envir)
> [1] "One"
>> objects(envir, all.names=TRUE) # is "Two" a 'hidden'
object?
> [1] "One" "Two"
>> exists("Two", envir=envir, inherits=FALSE)
> [1] TRUE
>> get("Two", envir=envir, inherits=FALSE) # get fails when
exists says ok
> Error in get("Two", envir = envir, inherits = FALSE) :
>    object 'Two' not found
>
> Should Rf_defineVar(sym, R_UnboundValue, envir) remove sym from envir?
>
> Bill Dunlap
> TIBCO Software
> wdunlap tibco.com
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel