Compiler (gcc) option to abort on nan?

[Joakim Hove]

Hello,

am developing a floating point intensive program with gcc. At some
point in the calculations i get 'nan' ascribed to a variable, this is
*wrong* (i.e. a bug in my code), however the program happily continues
execution, propagating the nan around.

Since I know that the occurrence of a 'nan' is an error (in my code
anyway), I would like the program to just dump core at that point. Is
it possible to compile/link the program in a way which achieves this?


Best Regards

Joakim Hove

 

[Anders Mikkelsen]

Hello,

am developing a floating point intensive program with gcc. At some
point in the calculations i get 'nan' ascribed to a variable, this is
*wrong* (i.e. a bug in my code), however the program happily continues
execution, propagating the nan around.

Since I know that the occurrence of a 'nan' is an error (in my code
anyway), I would like the program to just dump core at that point. Is
it possible to compile/link the program in a way which achieves this?

man assert


regards,
Anders

 

[Dan Pop]

am developing a floating point intensive program with gcc. At some
point in the calculations i get 'nan' ascribed to a variable, this is
*wrong* (i.e. a bug in my code), however the program happily continues
execution, propagating the nan around.

Since I know that the occurrence of a 'nan' is an error (in my code
anyway), I would like the program to just dump core at that point. Is
it possible to compile/link the program in a way which achieves this?

Ever considered grepping for "NaN" in the gcc man page?

Dan

 

[Brian Gough]

Since I know that the occurrence of a 'nan' is an error (in my code
anyway), I would like the program to just dump core at that point. Is
it possible to compile/link the program in a way which achieves this?

Enable and trap the signal SIGFPE.
(see e.g. http://www.gnu.org/software/gsl/manual/gsl-ref_39.html#SEC505)