Refactor for the meta functions #1332
Conversation
library methods or coalesce them to be more consistent
SteveBronder
changed the title
…gs/RELEASE_500/final)
…gs/RELEASE_500/final)
…gs/RELEASE_500/final)
…stable/2017-11-14)
This reverts commit a39fa30.
…stable/2017-11-14)
|
This seems like it will go green. If you dont have anyone else in mind for review feel free to assign me. I have enough experience with the traits in Stan Math that I feel comfortable with this. |
|
Awesome ty! Yeah I'll assign you if you don't mind |
| T, std::enable_if_t<std::is_const<T>::value || std::is_pointer<T>::value>> { | ||
| using type = typename scalar_type_base< | ||
| std::remove_pointer_t<std::remove_cv_t<T>>>::type; | ||
| }; |
There was a problem hiding this comment.
@bob-carpenter @syclik this is probs the only serious change to point out. I'm assuming the intent of scalar_type is to get the underlying value so this also cycles through pointers. Does this seem like it's right?
|
(stat_comp_benchmarks/benchmarks/gp_pois_regr/gp_pois_regr.stan, 0.98) |
…gs/RELEASE_500/final)
…stable/2017-11-14)
…stable/2017-11-14)
|
Edit: Nvm we are good!
|
|
(stat_comp_benchmarks/benchmarks/gp_pois_regr/gp_pois_regr.stan, 0.97) |
|
Glad this was solved, I will look into it to see why this happened. |
|
@SteveBronder will have a look this evening or tomorrow morning. |
|
(stat_comp_benchmarks/benchmarks/gp_pois_regr/gp_pois_regr.stan, 0.96) |
|
Awesome! I glanced at it and it looks rad. (@rok-cesnovar, stay on as the reviewer please. Thank you!) |
|
Sorry Steve, I havent gotten to this yet. Will do tmrw. |
|
No worries! |
…includes from from scal and arr tests. Fixed docs for is_var.hpp
Summary
This refactors the metaprogramming structs and aliases to be a bit more consistent in their definitions and use more of the C++14 standard library methods. A few of the highlights are
Instead of using partial specializations that required explicit specializations for const and & types, we use SFINAE and
enable_if_t<std::decay_t<T>>to turn on and off the specializations for things likeis_vectorEdit:
scalar_typeremoves const but leaves refs and pointers which is what the original implementation didscalar_seq_viewnow uses SFINAE to make sure types use the correct specialization.Deletes ad_promotable in fwd and rev. Instead we use the
std::is_convertiblecheck that the types are implicitly convertible.is_fvar depended on knowing fvar, but the
.hppof fvar depends on the meta folder which defined fvar! So I madefvar.hppnot depend on the meta programming.new structs
is_std_vectorandis_eigento check if something is a standard vector or derived from EigenBaseis_vectornow checks if the eigen vector has rows or columns at compile time equal to 1.Moved a number of metaprogramming from arr and mat into scale when it could be instantiated there with specializations for
vectorandeigenleft inarrandmatCleaned up some of the docs
The new struct
bool_constant, which is juststd::integral_constant<bool, bool B>, is inherited by most of the structs that are used to check for types at compile time.Some of the tests checked types for
std::vector<const double>orstd::vector<double&>which is illegal under C++11 (types in std::vectors have to be movable and erasable so these were changed to look for pointers.structs containing a type now use
usinginstead oftypedef13. constexpr helper variable templates so we can call(THX WINDOWS)is_vector_v<T>instead ofis_vector<T>::valueTests
The only thing I still need to do is update the tests for
is_std_vectorandis_eigenSide Effects
Should be none
Checklist
Math issue Update internals to use more modern c++ #1308
Copyright holder: Steve Bronder
the basic tests are passing
./runTests.py test/unit)make test-headers)make doxygen)make cpplint)(YAS) the code is written in idiomatic C++ and changes are documented in the doxygen
the new changes are tested