New user self-registration is disabled due to spam. Please email eigen-core-team @ lists.tuxfamily.org if you need an account.
Before reporting a bug, please make sure that your Eigen version is up-to-date!
Bug 1209 - No conversion from Eigen::Translation or Eigen::Rotation2D to Eigen::Transform
Summary: No conversion from Eigen::Translation or Eigen::Rotation2D to Eigen::Transform
Status: DECISIONNEEDED
Alias: None
Product: Eigen
Classification: Unclassified
Component: Geometry (show other bugs)
Version: 3.2
Hardware: All All
: Normal Compilation Problem
Assignee: Nobody
URL:
Whiteboard:
Keywords:
Depends on:
Blocks: 3.4
  Show dependency treegraph
 
Reported: 2016-04-22 15:31 UTC by sogartary
Modified: 2016-05-18 12:06 UTC (History)
4 users (show)



Attachments
A short example reproducing the bug (606 bytes, text/x-c++src)
2016-04-22 15:31 UTC, sogartary
no flags Details

Description sogartary 2016-04-22 15:31:05 UTC
Created attachment 700 [details]
A short example reproducing the bug

The documentation https://eigen.tuxfamily.org/dox/group__TutorialGeometry.html states:

"Any of the above transformation types can be converted to any other types of the same nature, or to a more generic type."

It seems there is no conversion from
Translation<float, 2> or Eigen::Rotation2D<float> to
Transform<float, 2, Eigen::Affine>

Attached is a file with a short code piece illustrating the problem.

The expected behavior is the program to compile without errors.

I have not checked for other types.
Comment 1 Christoph Hertzberg 2016-04-25 10:59:42 UTC
The documentation says that you need to convert from Rotation/Translation to Transform explicitly or by assignment, i.e.
  Eigen::Transform<float, 2, Eigen::Affine> t2 (Eigen::Rotation2D<float>(0));
or
  Eigen::Transform<float, 2, Eigen::Affine> t2; t2 = Eigen::Rotation2D<float>(0);

I don't know the rational why we prohibit implicit conversion at this point? 
IMO, if a class (such as Rotation) is conceptually a subclass of another (in this case Transform), I would expect that they can be converted implicitly. Generally, if there is an assignment operator from A to B, I'd usually expect to have an implicit constructor as well (unless the assigned object is some sort of delegate, such as an Eigen::Map, of course).

The behaviour is the same in 3.2 and 3.3 at the moment.
Comment 2 Gael Guennebaud 2016-04-25 13:44:30 UTC
This issue is reported over and over (like the missing template keyword in method calls).

I guess the main reason is to prevent costly conversion/temporary to occur silently in function calls.

The fact that in c++ T a = b; involves an explicit conversion from type of b to T is really annoying and an unwelcomed side effect.
Comment 3 Christoph Hertzberg 2016-04-25 14:17:18 UTC
I just dug out the discussion we had on that half a decade ago:
http://thread.gmane.org/gmane.comp.lib.eigen/2192

Basically, the question is whether explicit constructors shall only prevent programming errors (e.g., prohibit implicit conversion from size_t to VectorXd), or also help against introducing overhead. I do remember the motivation now -- and there actually is an easy workaround for users who do want to have explicit conversion. I'm still not entirely convinced that this necessarily makes sense (from a user's point of view). E.g.:

void foo(const Transform2d&);
void bar(){
  foo(Translation2d() * Rotation2d()); // works.
  foo(Translation2d());                // does not work!
  foo(Transform2d(Translation2d()) );  // unnecessarily verbose ...
}


Also, in other cases we have no problem with implicit conversion, e.g.:

void foo(const MatrixXd& mat) { }
void bar() {
   ArrayXXd arr;
   foo(arr); // expensive copy of arr into a MatrixXd;
}

Note You need to log in before you can comment on or make changes to this bug.