Skip to content

R C api for 'inherits' S3 and S4 objects

8 messages · Jan Gorecki, Luke Tierney, Gábor Csárdi +2 more

Original
Jan Gorecki
# 
Dear R developers,

Motivated by discussion about checking inheritance of S3 and S4
objects (in head matrix/array topic) I would light to shed some light
on a minor gap about that matter in R C API.
Currently we are able to check inheritance for S3 class objects from C
in a robust way (no allocation, thread safe). This is unfortunately
not possible for S4 classes. I would kindly request new function in R
C api so it can be achieved for S4 classes with no risk of allocation.
For reference mentioned functions below. Thank you.
Jan Gorecki

// S3 inheritance
bool INHERITS(SEXP x, SEXP char_) {
  SEXP klass;
  if (isString(klass = getAttrib(x, R_ClassSymbol))) {
    for (int i=0; i<LENGTH(klass); i++) {
      if (STRING_ELT(klass, i) == char_) return true;
    }
  }
  return false;
}
// S4 inheritance
bool Rinherits(SEXP x, SEXP char_) {
  SEXP vec = PROTECT(ScalarString(char_));
  SEXP call = PROTECT(lang3(sym_inherits, x, vec));
  bool ans = LOGICAL(eval(call, R_GlobalEnv))[0]==1;
  UNPROTECT(2);
  return ans;
}
Luke Tierney
#
On Fri, 1 Nov 2019, Jan Gorecki wrote:

            

      
      
      
    

        
Your premise is not correct. Rf_inherits will not GC but it can
allocate and is not thread safe.

Best,

luke

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Jan Gorecki
    

    

      
      #
    

  


  

      
Thank you Luke.
That is why I don't use Rf_inherits but INHERITS which does not
allocate, provided in the email body.
I cannot do similarly for S4 classes, thus asking for some API for that.

        
On Fri, Nov 1, 2019 at 5:56 PM Tierney, Luke <luke-tierney at uiowa.edu> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Gábor Csárdi
    

    

      
      #
    

  


  

      
AFAIR getAttrib() can allocate as well. Also, R API functions that do
not allocate today, may allocate in the future.

Gabor

        
On Fri, Nov 1, 2019 at 1:24 PM Jan Gorecki <j.gorecki at wit.edu.pl> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Simon Urbanek
    

    

      
      #
    

  


  

      
Note that your desire is by definition impossible - as your example also shows checking for S4 inheritance involves evaluation and thus allocation which cannot be avoided by the dynamic design of S4 inheritance.

Cheers,
Simon

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Luke Tierney
    

    

      
      #
    

  


  

      
On Fri, 1 Nov 2019, Jan Gorecki wrote:

        

            

      
      
      
    

        
Your definition can allocate because STING_ELT can allocate.
getAttrib can GC in general. Currently it would not GC or allocate in
this case, but this could change.

You can't assume thread-safety for calls into the R API, or any API
for that matter, unless they are documented to be thread-safe.

You would be better off using Rf_inherits as it does not make the
assumption that you can use pointer comparisons to check for identical
strings.  CHARSXPs are almost always cached but they are not
guaranteed to be, and the caching strategy might change in the future.

Best,

luke

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Michael Lawrence
    

    

      
      #
    

  


  

      
If your goal is to perform multithreaded computations, why not perform
all necessary interactions with R upfront and then compute only on
primitives? It would help for us to understand your use case.

        
On Fri, Nov 1, 2019 at 4:26 AM Jan Gorecki <j.gorecki at wit.edu.pl> wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Jan Gorecki
    

    

      
      #
    

  


  

      
Thank you all for your valuable comments.
Best,
Jan

        
On Fri, Nov 1, 2019 at 8:15 PM Tierney, Luke <luke-tierney at uiowa.edu> wrote: