[Rcpp-devel] Rcpp catching a domain error

[Devin]

Thanks Qiang for your help

When I try your code, this is the output I get in Rstudio:

status = 1.#QNAN0
status = 0.310474
[1] 0

What I try to achieve is to feed this function with different input parameters and everytime the function results in "1.#QNAN0", it should return "NAN".

I tried something like:

int test()
{
    gsl_set_error_handler_off();
    if(gsl_sf_hyperg_2F1(1,1,1.1467003,1) == "1.#QNAN0"){
        std::cout << "Error" << std::endl;
    }
    return 0;
}

... but you can't set a mathematical operation equal to a string. I also tried:

int test()
{
    gsl_set_error_handler_off();
    if(gsl_sf_hyperg_2F1(1,1,1.1467003,1) == NAN){
        std::cout << "Error" << std::endl;
    }
    return 0;
}

... but the if-branch gets ignored. How can I achieve that the program runs through the if-branch?

I appreciate your input!
Devin

-----Original Message-----
From: Dirk Eddelbuettel [mailto:edd at debian.org] 
Sent: Thursday, December 11, 2014 4:36 PM
To: Qiang Kou
Cc: Devin; rcpp-devel at lists.r-forge.r-project.org
Subject: Re: [Rcpp-devel] Rcpp catching a domain error


Hi there,

On 11 December 2014 at 10:27, Qiang Kou wrote:
| Hi, Devin,
| 
| I don't whether this is the standard solution.
| 
| As I understand, the default error handler in GSL will abort the 
| program after printing error message. So you need to close it first.

Qiang beat me to it, but when I ran your example (in a standard R session, in my case inside Emacs) I get 

  R> sourceCpp("/tmp/rcppgslabort.cpp")
  R> simpletest()
  gsl: hyperg_2F1.c:667: ERROR: domain error
  Default GSL error handler invoked.

  Process R aborted (core dumped) at Thu Dec 11 09:30:40 2014

which unmistakenly shows __GSL__ aborting here so our try/catch can't do anything for us.

Luckily Qiang is much on the case than I was:
 
| In most time I don't use try-catch with GSL, I detect whether the 
| result is 'nan' or not. Try the code below:
| 
| Best,
| 
| KK
| 
| // [[Rcpp::depends(RcppGSL)]]
| #include <RcppGSL.h>
| #include <gsl/gsl_sf_hyperg.h>
|  
| #include <Rcpp.h>
| #include <math.h>
| using namespace Rcpp;
|  
| // [[Rcpp::export]]
| int test()
| {
|     gsl_set_error_handler_off();
|     printf ("status = %f\n", gsl_sf_hyperg_2F1(1,1,1.1467003,1));
|     printf ("status = %f\n", 
| gsl_sf_hyperg_2F1(-10.34,2.05,3.05,0.1725));
|     return 0;
| }

That's a nice trick:

R> sourceCpp("/tmp/rcppgslabort.cpp")
R> fixedtest()
status = nan
status = 0.310474
[1] 0
R> 

Qiang, how about wrapping ten lines of context around it and making it an Rcpp Gallery post?  This ought to be documented....

Dirk

| 
| 
| On Thu, Dec 11, 2014 at 10:03 AM, Devin <devin.heer at gmail.com> wrote:
| 
| 
|     Dear list
| 
|      
| 
|     My Rcpp source file contains a gsl_sf_hyperg_2F1() function, and for some
|     input values to this function the return value is “NaN”, which leads in
|     RStudio to an abortion error.
| 
|      
| 
|     In order to avoid abortion errors, I try to implement a try/catch exception
|     handler:
| 
|    
|     // [[Rcpp::depends(RcppGSL)]]
| 
|     #include <RcppGSL.h>
| 
|     #include <gsl/gsl_sf_hyperg.h>
| 
|      
| 
|     #include <Rcpp.h>
| 
|     #include <math.h>
| 
|     using namespace Rcpp;
| 
|      
| 
|     // [[Rcpp::export]]
| 
|     int test()
| 
|     {
| 
|         try{
| 
|             gsl_sf_hyperg_2F1(1,1,1.1467003,1); //  abortion error since not
|     defined
| 
|         }
| 
|         catch(std::domain_error& e){
| 
|             std::cout << "Domain Error occurred!"  << std::endl;
| 
|         }
| 
|         return 0;
| 
|     }
| 
|      
| 
|     This seems to be wrong, since it still causes an abortion error in RStudio.
| 
|      
| 
|     I appreciate any input
| 
|      
| 
|     Thanks
| 
|     Devin
| 
|      
| 
|    
|     _______________________________________________
|     Rcpp-devel mailing list
|     Rcpp-devel at lists.r-forge.r-project.org
|     
| https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-deve
| l
| 
| 
| 
| 
| --
| Qiang Kou
| qkou at umail.iu.edu
| School of Informatics and Computing, Indiana University
| 
| _______________________________________________
| Rcpp-devel mailing list
| Rcpp-devel at lists.r-forge.r-project.org
| https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-deve
| l

--
http://dirk.eddelbuettel.com | @eddelbuettel | edd at debian.org