Re: [eigen] Re: "overloaded operator '*' is ambiguous..." in Eigen 3.3 ffa8356

[Gael Guennebaud <gael.guennebaud@xxxxxxxxx>]

Hi,

a simpler failing example is:

Eigen::Matrix<stan::math::fvar<double>, Eigen::Dynamic, Eigen::Dynamic> m(2, 2);

m * m;

the problem is that stan::math::fvar exposes a generic implicit constructor making it as if it can be converted from everything. More precisely, std::is_convertible<X,fvar> thus returns true for any type X, including an Eigen::Matrix<fvar>. In m * m, one of the factor can thus be interpreted as a compatible scalar.

This has to be fixed in fvar, for instance using SFINAE to enable the generic ctor only for valid types.

gael

On Mon, Oct 31, 2016 at 4:09 PM, Daniel Lee <bearlee@xxxxxxxxxxxx> wrote:

And thanks! I accidentally hit send before finishing.

Any help making it more minimal or if you know how operators have changed would probably be enough for me to create a simpler example.

Daniel

On Mon, Oct 31, 2016 at 11:08 AM, Daniel Lee <bearlee@xxxxxxxxxxxx> wrote:

Hi all,

First, thanks Gael for fixing the last bug! I missed the notification from the list when you first posted it..

On to the current bug. I want to apologize: I don't have a minimal working example demonstrating this issue. It would help me to figure out how to trap / create a minimal example out of this for future reference.

I'm using the Stan Math library (https://github.com/stan-dev/math.git) on this branch: feature/0187-eigen-3.3

The example I'm working with right now is this and it fails to compile (if you put this in test/tmp_test.cpp, it's easy to run using: ./runTests.py test/tmp_test.cpp).

#include <stan/math/fwd/mat.hpp>

#include <gtest/gtest.h>

TEST(int, main) {

  Eigen::Matrix<stan::math::fvar<double>, Eigen::Dynamic, Eigen::Dynamic> m(2, 2);

  m << 4, 1, 1, 4;

  Eigen::LLT<Eigen::Matrix<stan::math::fvar<double>, Eigen::Dynamic, Eigen::Dynamic> >

    llt(m.rows());

  llt.compute(m);

}

That last line, llt.compute(m), creates a slew of compiler messages. Any suggestions on how to trap this? This code works with Eigen 3.2.9. The only things that changed within the Stan Math library pertaining to the type stan::math::fvar<double> is the implementation of static int digits10() for printing.

clang++ -I . -isystem lib/eigen_3.3.ffa8356 -isystem lib/boost_1.60.0 -isystemlib/cvodes_2.9.0/include -Wall -DBOOST_RESULT_OF_USE_TR1 -DBOOST_NO_DECLTYPE -DBOOST_DISABLE_ASSERTS -DNO_FPRINTF_OUTPUT -pipe -I/usr/local/opt/llvm/include -Wno-unknown-warning-option -Wno-unused-function -Wno-tautological-compare -Wno-c++11-long-long -Wsign-compare -Wno-unused-local-typedef -ftemplate-depth=256   -c -O3 -DGTEST_USE_OWN_TR1_TUPLE -Wc++11-extensions -Wno-c++11-long-long -isystem lib/gtest_1.7.0/include -isystem lib/gtest_1.7.0 test/tmp_test.cpp -o test/tmp_test.o

In file included from test/tmp_test.cpp:2:

In file included from ./stan/math/fwd/mat.hpp:8:

In file included from ./stan/math/fwd/mat/vectorize/apply_scalar_unary.hpp:4:

In file included from ./stan/math/prim/mat/vectorize/apply_scalar_unary.hpp:4:

In file included from lib/eigen_3.3.ffa8356/Eigen/Dense:3:

In file included from lib/eigen_3.3.ffa8356/Eigen/Cholesky:31:

lib/eigen_3.3.ffa8356/Eigen/src/Cholesky/LLT.h:321:45: error: use of overloaded operator '*' is

      ambiguous (with operand types 'Block<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1,

      -1>, Dynamic, Dynamic>' and 'const AdjointReturnType' (aka 'const Eigen::Transpose<const

      Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1>, 1, -1, false> >'))

      if (k>0 && rs>0) A21.noalias() -= A20 * A10.adjoint();

                                        ~~~ ^ ~~~~~~~~~~~~~

lib/eigen_3.3.ffa8356/Eigen/src/Cholesky/LLT.h:333:14: note: in instantiation of function template

      specialization 'Eigen::internal::llt_inplace<stan::math::fvar<double>,

      1>::unblocked<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1> >' requested here

      return unblocked(m);

             ^

lib/eigen_3.3.ffa8356/Eigen/src/Cholesky/LLT.h:397:61: note: in instantiation of function template

      specialization 'Eigen::internal::llt_inplace<stan::math::fvar<double>,

      1>::blocked<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1> >' requested here

  { return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; }

                                                            ^

lib/eigen_3.3.ffa8356/Eigen/src/Cholesky/LLT.h:444:21: note: in instantiation of member function

      'Eigen::internal::LLT_Traits<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1>,

      1>::inplace_decomposition' requested here

  bool ok = Traits::inplace_decomposition(m_matrix);

                    ^

test/tmp_test.cpp:12:7: note: in instantiation of function template specialization

      'Eigen::LLT<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1>,

      1>::compute<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1> >' requested here

  llt.compute(m);

      ^

lib/eigen_3.3.ffa8356/Eigen/src/SparseCore/../plugins/CommonCwiseBinaryOps.h:50:29: note: candidate

      function [with T = Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1>, -1,

      -1, false>]

EIGEN_MAKE_SCALAR_BINARY_OP(operator*,product)

                            ^

lib/eigen_3.3.ffa8356/Eigen/src/SparseCore/../plugins/CommonCwiseBinaryOps.h:50:29: note: candidate

      function [with T = Eigen::Transpose<const Eigen::Block<Eigen::Matrix<stan::math::fvar<double>,

      -1, -1, 0, -1, -1>, 1, -1, false> >]

lib/eigen_3.3.ffa8356/Eigen/src/Core/MatrixBase.h:173:5: note: candidate function [with OtherDerived

      = Eigen::Transpose<const Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1,

      -1>, 1, -1, false> >]

    operator*(const MatrixBase<OtherDerived> &other) const;

    ^

In file included from test/tmp_test.cpp:2:

In file included from ./stan/math/fwd/mat.hpp:8:

In file included from ./stan/math/fwd/mat/vectorize/apply_scalar_unary.hpp:4:

In file included from ./stan/math/prim/mat/vectorize/apply_scalar_unary.hpp:4:

In file included from lib/eigen_3.3.ffa8356/Eigen/Dense:3:

In file included from lib/eigen_3.3.ffa8356/Eigen/Cholesky:31:

lib/eigen_3.3.ffa8356/Eigen/src/Cholesky/LLT.h:321:45: error: use of overloaded operator '*' is

      ambiguous (with operand types 'Block<Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1,

      -1, 0, -1, -1>, -1, -1, false>, Dynamic, Dynamic>' and 'const AdjointReturnType' (aka 'const

      Eigen::Transpose<const Eigen::Block<Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1,

      -1, 0, -1, -1>, -1, -1, false>, 1, -1, false> >'))

      if (k>0 && rs>0) A21.noalias() -= A20 * A10.adjoint();

                                        ~~~ ^ ~~~~~~~~~~~~~

lib/eigen_3.3.ffa8356/Eigen/src/Cholesky/LLT.h:352:15: note: in instantiation of function template

      specialization 'Eigen::internal::llt_inplace<stan::math::fvar<double>,

      1>::unblocked<Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1>, -1, -1,

      false> >' requested here

      if((ret=unblocked(A11))>=0) return k+ret;

              ^

lib/eigen_3.3.ffa8356/Eigen/src/Cholesky/LLT.h:397:61: note: in instantiation of function template

      specialization 'Eigen::internal::llt_inplace<stan::math::fvar<double>,

      1>::blocked<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1> >' requested here

  { return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; }

                                                            ^

lib/eigen_3.3.ffa8356/Eigen/src/Cholesky/LLT.h:444:21: note: in instantiation of member function

      'Eigen::internal::LLT_Traits<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1>,

      1>::inplace_decomposition' requested here

  bool ok = Traits::inplace_decomposition(m_matrix);

                    ^

test/tmp_test.cpp:12:7: note: in instantiation of function template specialization

      'Eigen::LLT<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1>,

      1>::compute<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1> >' requested here

  llt.compute(m);

      ^

lib/eigen_3.3.ffa8356/Eigen/src/SparseCore/../plugins/CommonCwiseBinaryOps.h:50:29: note: candidate

      function [with T = Eigen::Block<Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0,

      -1, -1>, -1, -1, false>, -1, -1, false>]

EIGEN_MAKE_SCALAR_BINARY_OP(operator*,product)

                            ^

lib/eigen_3.3.ffa8356/Eigen/src/SparseCore/../plugins/CommonCwiseBinaryOps.h:50:29: note: candidate

      function [with T = Eigen::Transpose<const

      Eigen::Block<Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1, -1, 0, -1, -1>, -1, -1,

      false>, 1, -1, false> >]

lib/eigen_3.3.ffa8356/Eigen/src/Core/MatrixBase.h:173:5: note: candidate function [with OtherDerived

      = Eigen::Transpose<const Eigen::Block<Eigen::Block<Eigen::Matrix<stan::math::fvar<double>, -1,

      -1, 0, -1, -1>, -1, -1, false>, 1, -1, false> >]

    operator*(const MatrixBase<OtherDerived> &other) const;

    ^

2 errors generated.

make: *** [test/tmp_test.o] Error 1

make test/tmp_test failed

exit now (10/31/16 11:04:33 EDT)