Skip to content

Fix ambiguous sqrt() overload caused by ADL

Tyler Venessrequested to merge
calcmogul/eigen:fix-ambiguous-sqrt-overload-caused-by-adl into master

Here's the compiler error:

/home/tav/git/Sleipnir/build/_deps/eigen3-src/Eigen/src/Householder/Householder.h:82:16: error: call of overloaded ‘sqrt(boost::decimal::decimal64_t)’ is ambiguous
   82 |     beta = sqrt(numext::abs2(c0) + tailSqNorm);
      |            ~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/tav/git/Sleipnir/build/_deps/eigen3-src/Eigen/src/Householder/Householder.h:82:16: note: there are 2 candidates
In file included from /home/tav/git/Sleipnir/build/_deps/eigen3-src/Eigen/Core:198,
                 from /home/tav/git/Sleipnir/test/src/optimization/cart_pole_problem_test.cpp:8:
/home/tav/git/Sleipnir/build/_deps/eigen3-src/Eigen/src/Core/MathFunctions.h:1384:75: note: candidate 1: ‘typename Eigen::internal::sqrt_retval<typename Eigen::internal::global_math_functions_filtering_base<Scalar>::type>::type Eigen::numext::sqrt(const Scalar&) [with Scalar = boost::decimal::decimal64_t; typename Eigen::internal::sqrt_retval<typename Eigen::internal::global_math_functions_filtering_base<Scalar>::type>::type = boost::decimal::decimal64_t; typename Eigen::internal::global_math_functions_filtering_base<Scalar>::type = boost::decimal::decimal64_t]’
 1384 | EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE EIGEN_MATHFUNC_RETVAL(sqrt, Scalar) sqrt(const Scalar& x) {
      |                                                                           ^~~~
In file included from /home/tav/git/Sleipnir/build/_deps/decimal-src/include/boost/decimal/detail/cmath/ellint_1.hpp:16,
                 from /home/tav/git/Sleipnir/build/_deps/decimal-src/include/boost/decimal/cmath.hpp:18,
                 from /home/tav/git/Sleipnir/build/_deps/decimal-src/include/boost/decimal.hpp:33,
                 from /home/tav/git/Sleipnir/test/include/scalar_types_under_test.hpp:6,
                 from /home/tav/git/Sleipnir/test/src/optimization/cart_pole_problem_test.cpp:19:
/home/tav/git/Sleipnir/build/_deps/decimal-src/include/boost/decimal/detail/cmath/sqrt.hpp:167:16: note: candidate 2: ‘constexpr T boost::decimal::sqrt(T) requires  is_decimal_floating_point_v<T> [with T = decimal64_t]’
  167 | constexpr auto sqrt(const T val) noexcept
      |                ^~~~

Calling a function via its unqualified name invokes argument-dependent lookup. In this case, since using numext::sqrt; was used, both numext::sqrt() and boost::decimal::sqrt() participated in overload resolution. Since only numext::sqrt() was intended, the fix is to call that overload directly instead.

Merge request reports