Skip to content

[R-pkg-devel] New API in R-devel & minimum version of R

5 messages · Josiah Parry, Ben Bolker, Michael Chirico +1 more

Original
Josiah Parry
# 
I'm trying to work through some new WARNINGs that I am experiencing *only* in
R-devel.

 Found non-API calls to R: ?BODY?, ?CLOENV?, ?ENCLOS?, ?FORMALS?

Take FORMALS, for example. IIUC, we should now be using R_ClosureFormals
instead. However, my understanding is that this is available only in R 4.5
main trunk.

If migrating to the R_ClosureFormals, should the minimum version of R for
the package be recorded in the DESCRIPTION as R >= 4.5?

Additionally, I'm somewhat at a loss for how to pass CRAN checks for
R-release and R-devel while using only R_ClosureFormals to adhere to
R-devel WARNING.

Should old versions have an upper bound on the supported R version as well
as a lower one e.g. Depends may have R >=4.1.0 <4.5.0?

Thanks in advance for your advice / insight!

- Josiah
Ben Bolker
# 
FWIW  Rcpp handles this (for CLOENV) with an ifdef:

             #if (defined(R_VERSION) && R_VERSION >= R_Version(4,5,0))
             return R_ClosureEnv(fun);
             #else
             return CLOENV(fun);
             #endif

https://github.com/RcppCore/Rcpp/blob/257e1977cd6e251d0a3d691050ad43fa29cf9666/inst/include/Rcpp/Function.h#L102-L106
On 3/19/25 11:08, Josiah Parry wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Michael Chirico
    

    

      
      #
    

  


  

      
data.table is doing the same in a number of places, Ivan even went so far
as to pin down a specific SVN commit for those as well:

#if R_VERSION < R_Version(4, 5, 0) || R_SVN_REVISION < 86702
#  define isDataFrame(x) isFrame(x) // #6180
#endif

https://github.com/Rdatatable/data.table/blob/0909048017d35f8b8a910c74549650f5de5a74a9/src/data.table.h#L21-L23

Sneaking onto my soapbox to complain that it really seems incorrect that R
CMD check started complaining about these issues before the API-compliant
functions were included in an R release. I was really surprised to see
these functions were not included in the most recent past release, for
example.

Anyway, 4.5.0 is due for release in a few weeks so this will become moot
momentarily.

        
On Wed, Mar 19, 2025 at 8:19?AM Ben Bolker <bbolker at gmail.com> wrote:

        

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Dirk Eddelbuettel
    

    

      
      #
    

  


  

      
On 19 March 2025 at 11:16, Ben Bolker wrote:

        
|    FWIW  Rcpp handles this (for CLOENV) with an ifdef:
| 
|              #if (defined(R_VERSION) && R_VERSION >= R_Version(4,5,0))
|              return R_ClosureEnv(fun);
|              #else
|              return CLOENV(fun);
|              #endif
| 
| https://github.com/RcppCore/Rcpp/blob/257e1977cd6e251d0a3d691050ad43fa29cf9666/inst/include/Rcpp/Function.h#L102-L106

Beat me to it while I was away from my desk :-) We added that last week after
we good the nag from CRAN.

We are to follow the r-devel version R which is accessible e.g. 'as a
service' at winbuilder and macbuilder, at rhub (v2), and of course by
compiling locally (as I still do). Or by using one the available containers:
rocker/r-devel and rocker/drd are two I look after and they get updated
weekly.

Also note that when I made the update above I spotted a nice section in WRE
(where one 'of course' needs the r-devel version) which has what I quote
below. You could make use of that too.

Dirk


Quote from WRE follows

6.21.8 Some backports ?

This section lists backports of recently added definitions that can be used
in packages that need to be compiled under older versions of R that do not
yet contain these entry points.

    #if R_VERSION < R_Version(4, 4, 1)
    #define allocLang Rf_allocLang
    
    SEXP Rf_allocLang(int n)
    {
        if (n > 0)
              return LCONS(R_NilValue, Rf_allocList(n - 1));
        else
              return R_NilValue;
    }
    #endif
    
    #if R_VERSION < R_Version(4, 5, 0)
    # define isDataFrame(x) Rf_isFrame(x)
    # define R_ClosureFormals(x) FORMALS(x)
    # define R_ClosureEnv(x) CLOENV(x)
    # define R_ParentEnv(x) ENCLOS(x)
    
    SEXP R_mkClosure(SEXP formals, SEXP body, SEXP env)
    {
        SEXP fun = Rf_allocSExp(CLOSXP);
        SET_FORMALS(fun, formals);
        SET_BODY(fun, body);
        SET_CLOENV(fun, env);
        return fun;
    }
    
    void CLEAR_ATTRIB(SEXP x)
    {
        SET_ATTRIB(x, R_NilValue);
        SET_OBJECT(x, 0);
        UNSET_S4_OBJECT(x);
    }
    #endif

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Dirk Eddelbuettel
    

    

      
      #
    

  


  

      
On 19 March 2025 at 09:29, Michael Chirico wrote:

        
| data.table is doing the same in a number of places, Ivan even went so far
| as to pin down a specific SVN commit for those as well:
| 
| #if R_VERSION < R_Version(4, 5, 0) || R_SVN_REVISION < 86702
| #  define isDataFrame(x) isFrame(x) // #6180
| #endif
| 
| https://github.com/Rdatatable/data.table/blob/0909048017d35f8b8a910c74549650f5de5a74a9/src/data.table.h#L21-L23
| 
| Sneaking onto my soapbox to complain that it really seems incorrect that R
| CMD check started complaining about these issues before the API-compliant
| functions were included in an R release. I was really surprised to see
| these functions were not included in the most recent past release, for
| example.

Seconded.  

(Also I was wrong in them previous email: we learned from Ben as lme4 got the nag.)

Dirk