Separate absolute and relative temperature

[cstjean]

Following the discussion in #137, I separated absolute and relative temperature units, similar to @burnpanck's suggestion. It works well so far. Notable new test cases:

            @test uconvert(abs°F, 0abs°C) == 32abs°F
            @test uconvert(°F, 0°C) == 0°F
            @test 1.0abs°C + 45°F == 26.0abs°C
            @test 1abs°C - 1K == 0abs°C
            @test 40abs°C - 95abs°F == 5°C
            @test_throws DimensionError 1abs°C + 1abs°F

This PR is a proposal for discussion. If the general approach is acceptable, then we might find a less ugly name than 35abs°C. And of course: documentation.

Regarding binary arithmetic on absolute temperatures, we can either disallow promotion of two absolute temperatures in general, and explicitly allow some operations (-, ==, ...?) or allow all operations by default and disallow some (+, *, /, ...?) by throwing an exception. I did the latter in this PR, but I also tested the former and it works too.

Fix #137 and #72

[codecov]

Codecov Report

Merging #176 into master will decrease coverage by 0.13%.
The diff coverage is 100%.

@@            Coverage Diff            @@
##           master    #176      +/-   ##
=========================================
- Coverage   97.53%   97.4%   -0.14%     
=========================================
  Files          15      15              
  Lines         691     694       +3     
=========================================
+ Hits          674     676       +2     
- Misses         17      18       +1

Impacted Files	Coverage Δ
src/temperature.jl	94.11% <100%> (-5.89%)	⬇️
src/pkgdefaults.jl	100% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6e5c2a4...a16f0b7. Read the comment docs.

[cstjean]

Here's one issue:

julia> mean([1abs°C, 2abs°C])
ERROR: DimensionError: 1 abs°C and 2 abs°C are not dimensionally compatible.
Stacktrace:
 [1] +(::Quantity{Int64,Unitful.Dimensions{(Unitful.Dimension{:AbsTemperature}(1.

We can fix it manually, but it's a bit worrying.

[RuiRojo]

(A nobody giving his opinion)

I wouldn't pay much attention to the mean issue. It just happens that the general implementation of mean assumes you can add the stuff, while that's not a requirement for a mean to make sense. mean_v2(x) = mean(x .- x[1]) + x[1].

I understand the appeal to make the new weirdos in town (units with offsets) bear the prefix. But I think that it would introduce less silent mistakes to make the traditional names (°C, °F, K) default to the absolute ones which IIUC would error more often when misused (you can't add them, multiply, etc). As is,uconvert(°F, 0°C) returning 0°F smells like trouble. Maybe it's even worth it to have a short prefix for both relative and absolutes.

[ajkeller34]

@RuiRojo Nobody's a nobody, thanks for your interest! Not sure what to think yet regarding the unit names.

@cstjean Thanks for the PR. I think I agree with the general premise (still open to further discussion of unit names) but I don't think there should be a new dimension defined. The dimension of temperature ought to be the same regardless of absolute / relative scales. I imagine you were trying for a minimally invasive mechanism to identify absolute / relative scale from the type, but rather than compromise on dimensional correctness I think we could generalize the Units types to permit affine transformations. Perhaps an extra type parameter that could either be nothing (no affine transformation) or a value ("absolute zero" on the scale).

[cstjean]

Thank you for the comments. I agree that a general solution would be better.

The dimension of temperature ought to be the same regardless of absolute / relative scales.

Conceptually, I agree, but pragmatically, I can define functions with dimensions in the signature, like is_boiling(::Temperature) = t < 100°C. That function should not accept relative temperature, but I can't specify that if absolute and relative temperatures share the same dimension.

I think we could generalize the Units types to permit affine transformations. Perhaps an extra type parameter that could either be nothing (no affine transformation) or a value ("absolute zero" on the scale).

Sounds good.

The current PR is good enough for our purposes, so I don't know when/if I'll have time to push this further. If anyone wants to take a shot at this, please do.

[burnpanck]

This is exciting! I'm not even a Julia user yet, but it's great to see how the language allows library writers to offer tools that do the right thing™, without sacrificing expressiveness. And of course, kudos to you guys! Now if only arrays would start at zero...

[ajkeller34]

Closed in favor of #177, discussion can continue there.