* Demonstrate deduction guides for normal_distribution.
* Add missing test case.
* add class template argument deduction guides for distributions templated on real type - issue #754
* Remove no-arg tests in test_dist_deduction_guides.cpp - issue #754
GCC-8 and clang 6-8 were unhappy with the no-arg cases, which use the
default template arg, not the deduction guide, anyway.
* remove unused deduction guide for fisher_f - issue #754
Co-authored-by: jzmaddock <john@johnmaddock.co.uk>
* Color Maps
* Make color maps constexpr
[ci skip]
* Add newton fractal example
[ci skip]
* Remove some unused code.
* Make the color map base class generic in size
Fix naming convention
[ci skip]
* Begin documentation.
* Move helper functions from example into tools header
[ci skip]
* Update docs and remove non-ASCII characters from example
* Add image to docs
* Reduce size of virdis_newton_fractal from 1.31MB to 131KB
[ci skip]
* Add performance file
* Don't force linear complexity and fix CI failure for old clang versions
* Convert color_maps to free functions.
* Add missing header and remove constexpr test
* Convert tabs to spaces
[ci skip]
* Fix compile tests and make static constexpr uniform across data
* Add swatches to docs.
* Fix image links in docs
[ci skip]
Co-authored-by: Nick Thompson <nathompson7@protonmail.com>
Stop referencing boost::lexical_cast even in templates which aren't instantiated.
Fix missing macro definition in tr1.hpp.
Correct include order in some tests so we get consistent definitions for BOOST_HAS_FLOAT128.
Correct silly typo in unchecked_factorial.hpp.
Remove TEST_STD define in config.hpp as it needlessly breaks the TR1 tests.
Remove lexical_cast.hpp workaround file.
Correct #pragma in tr1.hpp.
This removes insignificant values at either end of the integral by way of a performance improvement.
Remove the check within the inner loop for possible early exit as we have already removed the range of values which are likely to be insignificant in value during the first 2 passes.
Recent changes to tanh_sinh alters the order in which values are summed: this can randomly change the found error for either better or worse, and so the 1eps tolerance here was over-optimistic.
