runtime differences with Eigen 3.3-rc1 + patch

[bgoodri]

The coming 3.3 release of Eigen has more changes than usual
http://eigen.tuxfamily.org/index.php?title=3.3
And Stan Math has a few build errors with it.

./stan/math/prim/mat/fun/LDLT_factor.hpp:104:24: error: no type named 'solve_retval' in namespace 'Eigen::internal'
./stan/math/prim/mat/fun/LDLT_factor.hpp:104:36: error: expected member name or ';' after declaration specifiers
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:127:66: error: template argument for template type parameter must be a type
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:157:42: error: template argument for template type parameter must be a type
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:201:59: error: template argument for template type parameter must be a type
./stan/math/rev/mat/fun/LDLT_factor.hpp:85:24: error: no type named 'solve_retval' in namespace 'Eigen::internal'
./stan/math/rev/mat/fun/LDLT_factor.hpp:85:36: error: expected member name or ';' after declaration specifiers

There was a bug fix in a branch that's not a release yet that builds, but gives wrong answers as @syclik indicated in #187 (comment)

[bob-carpenter]

It would be great to fix this in such a way it works with the current Eigen and the forthcoming 3.3 release.

I took a look at the traits, but this is going to take some digging. Are those multiply operations in the NumTraits file really going to be used internally by Eigen for multiplication? I didn't quite follow what they were doing there or which interface they were implementing.

[bgoodri]

Now Eigen is in beta. Math is still failing.

goodrich@T540p:/opt/stan/lib/stan_math_2.7.0$ grep -F "error:" /tmp/eigen_test.txt | sort | uniq
./stan/math/prim/mat/fun/LDLT_factor.hpp:104:24: error: no type named 'solve_retval' in namespace 'Eigen::internal'
./stan/math/prim/mat/fun/LDLT_factor.hpp:104:36: error: expected member name or ';' after declaration specifiers
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:128:66: error: template argument for template type parameter must be a In file included from <built-in>:325:
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:128:66: error: template argument for template type parameter must be a type
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:128:66: error:          typename LhsScalar, typename LhsMapper, int LhsStorageOrder, bool ConjugateLhs,
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:158:42: error: template argument for template typeIn file included from  parameter must be a type
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:158:42: error: template argument for template type parameter must be a type
./stan/math/rev/mat/fun/Eigen_NumTraits.hpp:202:59: error: template argument for template type parameter must be a type
./stan/math/rev/mat/fun/LDLT_factor.hpp:85:24: error: no type named 'solve_retval' in namespace 'Eigen::internal'
./stan/math/rev/mat/fun/LDLT_factor.hpp:85:36: error: expected member name or ';' after declaration specifiers

[bob-carpenter]

RcppEigen is still at 3.2.8, as of today. I'm going to update to that and see if I can get it working with the 3.3 beta, but if not, I'll just update to 3.2.8, for which I'm going to create a new issue.

[bgoodri]

3.2.8 should be fine for now. But 3.3.0 is going to introduce a lot of
changes, so we need to test it a lot before it is released, which we can't
do if it doesn't build anything Eigen-related.

On Wed, Mar 16, 2016 at 3:50 PM, Bob Carpenter notifications@github.com
wrote:

RcppEigen is still at 3.2.8, as of today. I'm going to update to that and
see if I can get it working with the 3.3 beta, but if not, I'll just update
to 3.2.8, for which I'm going to create a new issue.

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#187 (comment)