namespace CppAD about cppad

[bradbell]

ad_value is in the CppAD namespace and so the CppAD version of abs is being called; see
https://coin-or.github.io/CppAD/doc/unary_standard_math.htm

There is more going on here. In the CppAD namespace, the abs function has been extended to the base type; see
https://coin-or.github.io/CppAD/doc/base_std_math.htm

Perhaps it would have been better if CppAD had left abs alone and just used fabs, which it also defines for both cases.

[jcarpent]

Ok. I got it. I did have in mind that because abs and ad_value are inside the same CppAD namespace, then the compiler will then be able to use abs without any precise namespace.

This means that https://github.com/joaoleal/CppADCodeGen/blob/2f88701530fed57dcd1793ba769fbe8d53a9bee1/include/cppad/cg/math.hpp#L20
may lead to some trouble, because CppAD::cg::CG and the related math functions are in two different namespaces.

Thanks a lot for your quick reply!

[bradbell]

I think that that putting these definitions in the CppAD::cg namespace will work for code in the CppAD::cg name space and putting it in the CppAD namespace will work everywhere.

[jcarpent]

I tried to call for instance abs function on a CppAD::cg::CG<Scalar> object and it was not working when abs was only defined in CppAD namespace.
But if we define abs in CppAD::cg namespace, it perfectly works.

[bradbell]

Would it be possible for you to construct a very short example that demonstrates this just using CppAD ? I want to make sure I understand the exact problem.

[jcarpent]

Yes, of course. This is an example of what I tried to mention:

# include <cppad/cg.hpp>

int main()
{
  CppAD::AD<double> ad_value(-1.);
  abs(ad_value); // works perfectly
  
  CppAD::cg::CG<double> cg_value(-1.);
  abs(cg_value); // does not compile because abs<double>(const CppAD::cg::CG<double>&) is defined in namespace CppAD and not CppAD::cg
  return -1;
}

[bradbell]

The type CppAD::cg::CG is not a CppAD type.
On the other hand, abs should work for

typedef CppAD::cg::CG<double> Base;
CppAD::AD<Base> cg_value(-1);
abs(cg_value);

OH, there is a problem here. CppAD needs to compute the base type values, as well as the AD values, for each entry.

[jcarpent]

Yes, indeed.

[jcarpent]

Thanks a lot for your help. I fixed the bug related to my issues in joaoleal/CppADCodeGen#16