Re: [eigen] Re: Eigen 3.3-rc2: no type named 'Traits' in 'Eigen::internal::general_matrix_matrix_product...'

[Daniel Lee <bearlee@xxxxxxxxxxxx>]

Hi Gael,

Thank you! That is immensely helpful. I'll start simple and work my way through specializations. And I'll let you know if there are any specializations that are necessary with custom types. 

Daniel

On Nov 10, 2016, at 7:24 AM, Gael Guennebaud <gael.guennebaud@xxxxxxxxx> wrote:

On Tue, Nov 8, 2016 at 9:28 AM, <Daniel.Vollmer@xxxxxx> wrote:

I agree that working with custom types is a bit difficult at the moment (especially with the need to specialise in the Eigen::internal namespace).

In 3.3, if the scalar type is already STL compliant (regarding std::numeric_limits<>) then there is noting to do. For instance, one can use boost::multiprecision::number<XX,et_off> without any template specialization. There is only Eigen::NumTraits::dummy_precision that has to be re-defined for floating point numbers (needed for fuzzy comparisons only).

Then you can specialize more of Eigen::NumTraits for fine-tuning. In 3.3 you can also specialize Eigen::ScalarBinaryOpTraits to finely control mixing of different scalar types, but that's usually not required.

If you need more specializations (with 3.3), please let us know so that we can improve even more the automatic support of custom scalar types.

gael