[Rcpp-devel] Debugging intermittant vignette re-building failure

[Fellows, Ian]

Thank you Bill,

I have tried gctorture, and low and behold I got an error in some of my code (Yay!). I was able to pin the gctorture error to something small and reproducible. It turns out in a one place I created temporary SEXP variables (see the example below), and I guess these were not protected automatically by Rcpp. I’ve changed these to RObject objects and things seem to work fine, though I’ll have to do a bunch more builds to be sure.

This has gotten me a little worried about other places in the code where SEXPs pop up. I’ve now limited the use of SEXPs to function parameters and return values. This is mostly for transport of objects to and from R. Let’s say I have a method like

SEXP afunc(){
  List lis;
  // add things to list
  return lis;
}

My understanding is that afunc safe to use in a module. Is it also safe to call from other C++ code, or is lis open to garbage collection?



Thanks, 
Ian



---------

library(inline)
library(Rcpp)
inc <- '
using namespace Rcpp;
class TestObj {
public:
  TestObj(int i){}
  
  void set(SEXP robj){
        try{
            Language call("as.factor",robj);
            RObject facSexp = call.eval();
            Language call1("as.integer",facSexp);
            SEXP intSexp = call1.eval();
            Language call2("levels",facSexp);
            SEXP levelsSexp = call2.eval();
            vals = as< std::vector<int> >(intSexp); //here be errors with gctorture
        } catch(std::exception &ex) {		
            forward_exception_to_r(ex);
        } catch(...){
            ::Rf_error("error, invalid object addDiscreteVariableR");
        }
    
  }
  std::vector<int> vals;
};
RCPP_MODULE(mod) {
class_<TestObj>( "TestObj" )
.constructor<int>()
.method( "set", &TestObj::set )
;
}
'
fx <- cxxfunction(signature(), plugin="Rcpp", include=inc)
mod <- Module("mod", getDynLib(fx))
obj <- new(mod$TestObj, 1L)
vals <- 1:4
gc()
gctorture(T)
obj$set(vals)
# Error in obj$set(vals) : 
#   Not compatible with requested type: [type=list; target=integer].

--------





On January 8, 2019 at 2:35:30 PM, William Dunlap (wdunlap at tibco.com(mailto:wdunlap at tibco.com)) wrote:

> > --- re-building ‘lolog-ergm.Rmd’ using rmarkdown
> > Quitting from lines 69-75 (lolog-ergm.Rmd)
> > Error: processing vignette 'lolog-ergm.Rmd' failed with diagnostics:
> > unimplemented type (28) in 'duplicate'
>  
> I don't believe there is a type (*SXP) 28 in R. This could be due to memory misuse. Have you tried running the R code in lolog-ergm.Rmd (and your unit tests) in R  
> after setting gctorture(TRUE)?  
>  
> Bill Dunlap
> TIBCO Software
> wdunlap tibco.com(http://tibco.com)  
>  
> On Tue, Jan 8, 2019 at 1:31 PM Fellows, Ian wrote:
> >  
> > Hi All,  
> >  
> > I have a package ( https://github.com/statnet/lolog | https://cran.r-project.org/web/packages/lolog/index.html ) that makes extensive use of Rcpp modules and passes module objects up and down between R and C++ regularly. I’ve used the package extensively in large simulation studies, and analyses where it has been stable.
> >  
> > However, I have become aware of an intermittent failure to re-build the vignettes during R CMD check. I’ve only ever encountered the issue during check. Every once in a while I get WARNINGS like the following:
> >  
> >  
> > -------
> >  
> > * checking re-building of vignette outputs ... WARNING
> > Error(s) in re-building vignettes:
> > ...
> > --- re-building ‘lolog-ergm.Rmd’ using rmarkdown
> > Quitting from lines 69-75 (lolog-ergm.Rmd)
> > Error: processing vignette 'lolog-ergm.Rmd' failed with diagnostics:
> > unimplemented type (28) in 'duplicate'
> >  
> > --- failed re-building ‘lolog-ergm.Rmd’
> >  
> > --- re-building ‘lolog-introduction.Rmd’ using rmarkdown
> > --- finished re-building ‘lolog-introduction.Rmd’
> >  
> > SUMMARY: processing the following file failed:
> > ‘lolog-ergm.Rmd’
> >  
> > Error: Vignette re-building failed.
> > Execution halted
> >  
> > -------
> >  
> > or
> >  
> > -------
> >  
> > * checking re-building of vignette outputs ... [336s/452s] WARNING
> > Error(s) in re-building vignettes:
> > ...
> > --- re-building ‘lolog-ergm.Rmd’ using rmarkdown
> > --- finished re-building ‘lolog-ergm.Rmd’
> >  
> > --- re-building ‘lolog-introduction.Rmd’ using rmarkdown
> > Quitting from lines 355-381 (lolog-introduction.Rmd)  
> > Error: processing vignette 'lolog-introduction.Rmd' failed with diagnostics:
> > cannot coerce type 'environment' to vector of type 'any'
> > --- failed re-building ‘lolog-introduction.Rmd’
> >  
> > SUMMARY: processing the following file failed:
> > ‘lolog-introduction.Rmd’
> >  
> > Error: Vignette re-building failed.
> > Execution halted
> >  
> > -------
> >  
> >  
> > I’ve attempted to isolate the problem to something of a smaller but have thus far been unsuccessful. Since this only happens occasionally, I thought it might have something to do with how I’m passing objects up and down ( see for example: https://github.com/statnet/lolog/blob/master/inst/include/Model.h#L89 and https://github.com/statnet/lolog/blob/master/inst/include/util.h#L29 ), but that is just a shot in the dark.
> >  
> > Does anyone have any insight as to how i might be able to better debug this or intuitions about what could be the cause? Is it possible to get a stack trace when rebuilding the vignettes?
> >  
> > My apologies for the lack of a simple reproducible example.
> >  
> > Sincerely,
> > Ian Fellows
> >  
> >  
> >  
> >  
> >  
> >  
> > _______________________________________________
> > Rcpp-devel mailing list
> > Rcpp-devel at lists.r-forge.r-project.org(mailto:Rcpp-devel at lists.r-forge.r-project.org)
> > https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel