Named colors shouldn't be Tuple{Int64,Int64,Int64} about colors.jl

[timholy]

History-wise, I think these were added by people who are mostly doing HTML stuff and thus writing out JSON strings.

But we did eventually develop a more Julian approach:

julia> using Colors

julia> colorant"indianred"
RGB{N0f8}(0.804,0.361,0.361)

This actually does the work at compile-time, so performance-wise is equivalent to inserting the constant RGB{N0f8}(0.804,0.361,0.361). However, it requires an explicit color name; if you want to parse this for a variable, then use parse(Colorant, cname) where cname is a string.

I see the docs make this a bit confusing. Would you by any chance be willing to submit a pull request that clarifies this?

[scls19fr]

PR submitted.

but maybe this part of doc could be a good place to show how to convert to RGB{N0f32} (for example)

but it could also be in https://github.com/JuliaGraphics/ColorTypes.jl

[scls19fr]

RGB{N0f32}(colorant"indianred")

[timholy]

@scls19fr, why would convert to RGB{N0f32}? Do you realize that is 96-bit color? Did you actually mean to convert to RGB24 by any chance?

[scls19fr]

oh yes... that's a bit overkilled

[scls19fr]

Given values in tuples, I'm wondering if

https://github.com/JuliaGraphics/Colors.jl/blob/0f4c744902f22fb73776e50e839a2bf8ebfdc9d6/src/names_data.jl

shouldn't be changed from Int64 to UInt8

[timholy]

I'm less worried about the data type and more about the value; should we use 1.0 instead of 255?

The current conversions (including reinterpreting as N0f8) are done algorithmically here:

Colors.jl/src/parse.jl

Lines 108 to 110 in 33ce6c0

	return RGB{N0f8}(reinterpret(N0f8, UInt8(c[1])),
	reinterpret(N0f8, UInt8(c[2])),
	reinterpret(N0f8, UInt8(c[3])))

.

[scls19fr]

Working with float can lead to numerical issues...

this kind of values should be stored in a structure which can store for example fractional values (1/3 for example)

My 2 cts

[timholy]

Just FYI that table is only for internal use anyway, I don't think there's anything about it that leaks out to the user. By reinterpreting as N0f8 we're effectively storing all of them at ratios with denominator 255.

julia> colorant"indianred"
RGB{N0f8}(0.804,0.361,0.361)

julia> dump(colorant"indianred")
RGB{FixedPointNumbers.Normed{UInt8,8}}
  r: FixedPointNumbers.Normed{UInt8,8}
    i: UInt8 0xcd
  g: FixedPointNumbers.Normed{UInt8,8}
    i: UInt8 0x5c
  b: FixedPointNumbers.Normed{UInt8,8}
    i: UInt8 0x5c