v2 -> v3 porting

[Guillaume227]

I have been attempting to convert a codebase using version 2 to the 3.x branch and here is some feedback and questions.
I embarked on this upgrade because I grew tired of the inconsistent to_string vs stream behavior for units like volt_t.
My initial, naive, expectation was a drop-in replacement with no change needed in the rest of the code base.

Here is what happened in the hope that's useful for people going through an update. I am planning to send a few PRs as follow-up.

Most of the issues stemmed from using a float underlying type - all of the unit testing is done for double.

1. v2 Types are templates in v3
   I used to do:

#define UNIT_LIB_DEFAULT_TYPE float 
using meter_t = units::length::meter_t;

In 3.x I do:
using meter_t = units::length::meter_t<float>;

2. Dealing with literal operators and underlying type
   In 3..x it a bit less nicely packaged when it comes to literal operators since I am dealing with floats.
   I have to redefine them outside of core.h with a user-land macro since I do not want double:

 constexpr namespaceName::nameSingular##_t<float> operator""_##abbreviation(long double d) noexcept \
      { \
          return namespaceName::nameSingular##_t<float>(static_cast<float>(d)); \
      } 

It's not too bad though, I just had to reintroduce UNIT_HAS_LITERAL_SUPPORT in 3.x.

3. units::math

• It's easy enough to figure out that the math functions are directly available in the units namespace and that units::math is gone.
• In a more subtle way, I started to get compilation errors with units::sqrt of the no-template-overload-found variety.
  I tracked it down to a 0.5 literal that was causing a mixture of double and floats in the call to sqrtNewtonRaphson. Once found, it was straightforward to replace that with T{0.5}.

4. Ternary operator and arithmetic operators+,- return type
   I hit an issue where the ternary operator no longer compiles:

degree_t val1 = 10_deg;
degree_t val2 = 90_deg;
degree_t new_val = some_bool ? val1 -  val2 : val2;

After sifting through the template compilation errors, it's down to:
std::common_type_t<degree_t, degree_t> differing from degree_t, as it's using a dummy 1 conversion factor.
That can be solved through another std::common_type_t specialization.

[Guillaume227]

Wait, it seems I was missing some significant updates to v3.x, for example the _t suffix is gone entirely now. I need to review and update what I said above.

[Guillaume227]

Ok after a bit of renaming I managed to make almost everything work on 3.x - I updated all the PRs above, they should be good for review/feedback.
The one thing that's not working for me in 3.x as it used to on 2.3 is to_string for on-the-fly units, e.g. :
to_string(1_A * 1_A);
It used to print:
1 A^2
but now I am getting a compilation error.
Not a bit deal for me at the moment, I might look into that at a later stage unless someone has a better understand of what's going on here.

I also opened #318 and #317, they are self-explanatory

Overall things feel much cleaner in 3.x, Thanks. Straightening up the ADL story makes a big difference.

Regarding testing, my use case makes use of float underlying_type which makes me notice how current tests cover double underlying_type. I am not sure how to extend that without a lot of code duplication. Something like templating the test cases on underlying type?