atan2() should be in the same namespace as quantity so that ADL finds it.

[tjgolubi]

I was playing with my home-grown Vector and Point classes, trying to get them to work with mp-units. I wanted to deduce the angle type from the return type of atan2(y, x). If y and x are quantities, then the return type would by quantity<isq_angle[radian], double> (probably). However, ADL didn't find atan2() with quantityisq::displacement[m] arguments because atan2() and oher trig functions are in mp_units::angular (for some reason). What do you think about moving the trig functions into the mp_units namespace?

[mpusz]

Hi,

The problem here is that we have two atan2 functions:

• mp-units/src/systems/include/mp-units/systems/si/math.h

  Lines 118 to 121 in c9d7384

  	template<ReferenceOf<dimensionless> auto R, typename Rep>
  	requires requires(Rep v) { atan(v); } || requires(Rep v) { std::atan(v); }
  	[[nodiscard]] inline QuantityOf<MP_UNITS_IS_VALUE_WORKAROUND(isq::angular_measure)> auto atan(
  	const quantity<R, Rep>& q) noexcept

• mp-units/src/systems/include/mp-units/systems/angular/math.h

  Lines 116 to 118 in c9d7384

  	template<ReferenceOf<dimensionless> auto R, typename Rep>
  	requires requires(Rep v) { atan(v); } || requires(Rep v) { std::atan(v); }
  	[[nodiscard]] inline QuantityOf<angle> auto atan(const quantity<R, Rep>& q) noexcept

We can't put them in the same namespace as they differ only by the return type.

[tjgolubi]

It seems to me that the QuantityOf<angle> version of atan2() is sufficient and the QuantityOf<isq::angular_measure> is "extra/derived" and should require a user-initiated conversion either by context or explicitly.  I'm just a joe-user, but that's what I expected a quantity<angle[radian]> from atan2(quantity, quantity).  "radian" is obvious (over degrees, revolutions, grads) because "math" (d/dx sin(x) = cos(x) iff x is in radians).  "Angle" is the most basic angular type.  An angle isn't *really" dimensionless, is it?  I do use "isq::angular_measure" and others when I have time to figure out which is best, but that's my problem.  I "feel" that atan2(quantity, quantity) should return only have one return value in mp_units namespace.  I'm free to specialize my quantities and use my own namespace so that ADL atan2() gives the special return type that I want.  For example, I could call isq::atan2() when that's what I want. Angles are really fundamental to vectors, they have a magnitude *and* a direction.  ADL is really fundamental to generic programming.  Now is the time to figure this stuff out. Thanks for all the hard work.  I like the direction you're taking.  The documentation is also impressive. ⁣terry​

…

On Nov 2, 2025, 4:39 PM, at 4:39 PM, Mateusz Pusz ***@***.***> wrote: Hi, The problem here is that we have two `atan2` functions: - https://github.com/mpusz/mp-units/blob/c9d7384cc7130a6377e0e37d4c4344ff94988829/src/systems/include/mp-units/systems/si/math.h#L118-L121 - https://github.com/mpusz/mp-units/blob/c9d7384cc7130a6377e0e37d4c4344ff94988829/src/systems/include/mp-units/systems/angular/math.h#L116-L118 We can't put them in the same namespace as they differ only by the return type. -- Reply to this email directly or view it on GitHub: #737 (comment) You are receiving this because you authored the thread. Message ID: ***@***.***>

[mpusz]

Hi Terry!

It seems to me that the QuantityOf version of atan2() is sufficient and the QuantityOfisq::angular_measure is "extra/derived" and should require a user-initiated conversion either by context or explicitly.

Actually, it is not the case. Most of our users will use SI which provides the other overload. angular system is an experimental system of quantities and units that makes an angle a base quantity (rather than dimensionless). Most users do not use it and there is a high chance it won't be standardized.

Both of them return radian:

• si::radian
• angular::radian

An angle isn't *really" dimensionless, is it?

Unfortunately it is 😞 As long as ISQ and SI do not admit that angle has a base dimension/quantity/unit we need to have those two separate. If you insist on having only one, then we unfortunately use the one you are not interested in (from the SI system). Please read more here: https://mpusz.github.io/mp-units/latest/users_guide/systems/strong_angular_system/.

Please let me know your thoughts on this.

[tjgolubi]

Smarter people than I have been discussing this for a long time. I'd just want ADL to work. I don't care which mp_units::quantity<R, radians> mp_units::atan2(quantity, quantity) returns. Pick any one. Or is there a better way for my template Vector3D class to deduce the angle type other than using the return type atan2()?

Nice explanation of the problem. I didn't understand its relevance the first time I read it. The angular velocity = linear velocity / radius example demonstrates the issue very well.

[mpusz]

I'd just want ADL to work. ... Pick any one.

It is not that easy as you can see. We can't put both of them in the mp-units namespace because they do not form a valid overload set. Prioritizing one makes the other one much harder to use. Even if P2822 will be accepted for C++29, it could improve tan() but not atan.

Or is there a better way for my template Vector3D class to deduce the angle type other than using the return type atan2()?

I would recommend you providing a customization point in your project. A type trait or a variable template that you could specialize. You can check https://en.cppreference.com/w/cpp/iterator/incrementable_traits.html for inspiration.

I am sorry that I can't help more for now.