Rcpp: Environment.get() version throwing an exception instead of returning NULL

[progtw]

In R one can search an environment without inheritance and find names that have NULL assigned:
get( "varName", envir=enEnv, inherits=FALSE)

Rcpp::Environement::get method returns NULL if the name is not assigned. Hence it cannot distinguish between a name bound to NULL or a nonexisting name.

Rcpp::Environement::find methods throws an error if the name is not bound. Hence, it is suitable to distinguish unbound versus NULL-bound names. However it corresponds to inherits=TRUE.

Hence, a method in the Environment class would be welcomed that throws an exception instead of returning NULL but only searches the environment without its parents. Name suggestion: SEXP getThrows( const std::string& name) const

[eddelbuettel]

Good report. Would you mind just putting together a five or ten liner we can sourceCpp() to have concrete behaviour? A working example of what exactly you don R and what to do from C++ would also help.

The NULL as 'not here' as well explicit assigned NULL is a little unfortunate. We can probably extend the API to add a bool isSet(const std::string& varname) or something like that.

[kevinushey]

The other alternate is to return the sentinel R value R_UnboundValue, and instruct users to check against that value.

[progtw]

Here the code with concrete behaviour.

get_cpp <- cxxfunction(signature(name_x="character", envir_x="environment"),    
        plugin="Rcpp"
        ,body='
                std::string name = as<std::string>(name_x);
                Environment envir(envir_x);
                return envir.get(name);
                ')

find_cpp <- cxxfunction(signature(name_x="character", envir_x="environment"),   
        plugin="Rcpp"
        ,body='
                std::string name = as<std::string>(name_x);
                Environment envir(envir_x);
                return envir.find(name);
                ')

c <- "cParent"      # value in parent environment
a <- "aParent"
env <- new.env()
env$a <- NULL

stopifnot(  get("a", envir=env, inherits=FALSE) == NULL )
try(  get("c", envir=env, inherits=FALSE)  )    # expect error
stopifnot(  get("a", envir=env, inherits=TRUE) == NULL )

get_cpp("a", env)   # NULL wanted behaviour
get_cpp("nonExisting", env) # NULL instead of notification that it is not existing
get_cpp("c", env)   # NULL instead of notification that it is not existing

find_cpp("a", env)  # NULL wanted behaviour
try(find_cpp("nonExisting", env))  # error: wanted behaviour 
find_cpp("c", env)  # "cParent": instead of notification 

[eddelbuettel]

Same thing using Rcpp Attributes which makes it a lot more concide.

Use sourceCpp("nameOfTheFile.cpp") to _build and run the R illustration_ in one step.

#include <Rcpp.h>

using namespace Rcpp;

// [[Rcpp::export]]
SEXP get_cpp(std::string name, Environment envir) {
  return envir.get(name);
}

// [[Rcpp::export]]
SEXP find_cpp(std::string name, Environment envir) {   
  return envir.find(name);
}

/*** R
c <- "cParent"      # value in parent environment
a <- "aParent"
env <- new.env()
env$a <- NULL

stopifnot(  get("a", envir=env, inherits=FALSE) == NULL )
try(  get("c", envir=env, inherits=FALSE)  )    # expect error
stopifnot(  get("a", envir=env, inherits=TRUE) == NULL )

get_cpp("a", env)   # NULL wanted behaviour
get_cpp("nonExisting", env) # NULL instead of notification that it is not existing
get_cpp("c", env)   # NULL instead of notification that it is not existing

find_cpp("a", env)  # NULL wanted behaviour
try(find_cpp("nonExisting", env))  # error: wanted behaviour 
find_cpp("c", env)  # "cParent": instead of notification 
*/

Edit: Made the return value a SEXP as we do not know what is being looked up.

[progtw]

I am not familiar with R Internals. But being able to check the return value with an if-statement (as I understood @kevinushey) in order to distinguish unbound versus NULL values would probably be better and faster than using try-catch statements. An example of how to do this check should then be provided in the function documentation.

[coatless]

So, harvesting from Environment get the isSet is straightforward with the exception of what to do in regard to promises. Should the promise be evaluated and if there is a result then the value is set? Or is just the presence of the promise sufficient enough to trigger a set condition?

I've opted to go along the lines of the later here:

bool isSet(const std::string& name) const {
    SEXP env = Storage::get__() ;
    SEXP nameSym = Rf_install(name.c_str());
    SEXP res = Rf_findVarInFrame( env, nameSym ) ;

    if( res == R_UnboundValue ) return false; // not found in environment

    return true; // object _might_ be a promise in this case.
}

> is_cpp("a", env)   # exists
[1] TRUE

> is_cpp("nonExisting", env) #  not existing
[1] FALSE

> is_cpp("c", env)   #  not existing
[1] FALSE

[coatless]

@progtw, sorry for not seeing this earlier, but the above function is not necessary as Rcpp already has one that provides the same functionality via exists.

Example:

// [[Rcpp::export]]
bool exists_cpp(std::string name, Environment envir) {   
  return envir.exists(name);
}

Only in the case where a promise object needs to be correctly evaluated is a new function (or a parameter on exists) required.

[coatless]

@eddelbuettel: This issue can be closed out as the functionality already exists in the member function exists.