Remove integer conversion and arithmetic

[jakobnissen]

As discussed on Slack

Currently, BioSymbols can be converted to integers:

julia> convert(Integer, DNA_M)
0x03

The problem with this is that convert is called automatically by Julia in many circumstances, which can lead to unexpected behaviour:

julia> a = dna"AAA"; a[1] = 3; a
3nt DNA Sequence:
MAA

The fact that DNA_M is internally stored as 0x03 is somewhat arbitrary implementation detail and should not be a part of the API for BioSymbols or BioSequences. Similarly, it's not entirely sensical why

julia> DNA_M + 1
DNA_G

In this PR, integer/biosymbol conversion and arithmetic is removed. This includes addition, subtraction, binary operations (|, ~ and &), and conversion.

• To get an integer from a biosymbol x, use encoded_data(x).
• To get a biosymbol from an integer, use encode(T, x).
• Any future BioSymbols that might not be implemented as primitive types should implement:
  *encoded_data_eltype(::Type{T}),
  *encoded_data(x::T)
  *encode(::Type{T}, x)
  The two latter methods fall back to using reinterpret.

Edit: This is very breaking and will with almost 100% certainty create problems for other BioJulia packages. I can fix those, if we agree this should be merged

[codecov]

Codecov Report

Merging #38 into master will increase coverage by 1.58%.
The diff coverage is 100.00%.

@@             Coverage Diff             @@
##           master       #38      +/-   ##
===========================================
+ Coverage   98.41%   100.00%   +1.58%     
===========================================
  Files           3         3              
  Lines         126       112      -14     
===========================================
- Hits          124       112      -12     
+ Misses          2         0       -2     

Flag	Coverage Δ
#unittests	100.00% <100.00%> (+1.58%)	⬆️

Impacted Files	Coverage Δ
src/BioSymbols.jl	100.00% <100.00%> (+6.89%)	⬆️
src/aminoacid.jl	100.00% <100.00%> (ø)
src/nucleicacid.jl	100.00% <100.00%> (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 04ee588...ae96f92. Read the comment docs.

[jakobnissen]

Re-added logical operations for nucleotides. I guess it actually is useful and meaningful to have DNA_A | DNA_G == DNA_R.

[CiaranOMara]

Concerning BioJulia/BioSequences.jl#102, I think "encode" means the compression of the 8-bit primitive.

What about using a method structure like this instead, which neatly skips around the need to choose a name?

function (s::Type{BioSymbol})(bits::UInt8)
    return reinterpret(s, bits)
end

function (s::Type{BioSymbol})(char::Char)
    return ...
end

[TransGirlCodes]

Re-added logical operations for nucleotides. I guess it actually is useful and meaningful to have DNA_A | DNA_G == DNA_R.

Yeah, I think keeping the logical and bitwise ops is a good idea. The docs describe how the symbols are encoded - DNA being "one-hot", and the logical ops then make implementing a lot of things like complements and set memberships and so on a breeze.

[CiaranOMara]

@jakobnissen, apologies, today I realise it is the 8-bit symbol encoding that is compressed.

[jakobnissen]

Can we merge this, @benjward ?

[TransGirlCodes]

@jakobnissen I think this is fine, will be a major release though. I'm happy with use of encode and encoded_data here but would be as happy just with constructor methods. encode makes sense to us from BioSequences, but explicit constructor methods might make more sense to the non-specialist julia user. We can also of course have both and one just calls the other.