Final refactoring of alphabets

[h-2]

These are the discussed changes and some smaller additions by me not yet discussed.
@seqan/core please have a look and give me objections ASAP since I want to start working on this very soon.

seqan3::SemialphabetConcept:

• refines std::StrictTotallyOrdered (order required to respond to ranks)
• refines std::Copyable and requires std::is_trivially_copyable
• requirement for std::Regular is dropped, but recommended for all type definitions

• to_rank customisation point object
  • looks for member function
  • looks for free function via ADL
  • not picked up via ADL itself

• alphabet_size_v
  • looks for static member variable ::alphabet_size (now it's ::value_size – a weird name leftover from SeqAn2)
  • looks for constexpr free function via ADL alphabet_size() which makes it possible to define your own alphabet without overriding the metafunction in SeqAn

seqan3::WritableSemialphabetConcept:

• refines seqan3::SemialphabetConcept
• assign_rank_to(rank, alph) customisation point object
  • was assign_rank(alph, rank) in seqan2-(target, source)-style; new style is more intuitive and allows for assign_rank_to<dna5>(rank) instead of explicit construction assign_rank(dna5{}, rank) [also necessary because we don't require regularity anymore]
  • looks for assign_rank(rank) member function
  • looks for assign_rank_to(rank, alph) free function via ADL
  • not picked up via ADL itself
  • the CPO defines a generic overload for assign_rank_to(rank, alph &&) so this does not have to be customised (only the overload for lvalue-references has to be customised).

seqan3::AlphabetConcept

• refines seqan3::SemialphabetConcept
• to_char customisation point analogous to above

seqan3::WritableAlphabetConcept

• refines seqan3::AlphabetConcept and seqan3::WritableSemialphabetConcept
• assign_char_to(char, alph) customisation point object analogous to assign_rank_to(rank, alph)

PLUS Constexpr versions of all of the above in seqan3::detail::.

• Note that underlying_rank_t and underlying_char_t will be renamed to alphabet_rank_t and alphabet_char_t and will not be customisation points, they always resolve to the return type of to_char() and `to_rank().

• rename alphabet_size_v -> alphabet_size

• rename ::value_size -> alphabet_size

[eseiler]

In general I agree. I'm just wondering why we are dropping the std::Regular requirement. How will we adjust all the, e.g., unit tests where we require to default construct a alphabet?

[h-2]

We don't have to adapt our tests because our alphabets will still be Regular. We just don't require it.

The reason was, as discussed in escala, that default constructability is not required by interfaces that only consume alphabets and that it prevents references from modelling the concepts.

[smehringer]

Hm... didn't we have the function style guide for parameters (out, in/out, in)? the proposed changes to the free functions would be inconsistent with this..

[h-2]

Hm... didn't we have the function style guide for parameters (out, in/out, in)? the proposed changes to the free functions would be inconsistent with this..

No, that was SeqAn2-style. In SeqAn3 it's "natural order" and has been since day 1:
https://github.com/seqan/seqan3/wiki/Functions#function-argument-number-and-order

[smehringer]

Then I'm fine with the changes

[eseiler]

We don't have to adapt our tests because our alphabets will still be Regular. We just don't require it.

So we implicitly require it for our alphabets then.

The reason was, as discussed in escala, that default constructability is not required by interfaces that only consume alphabets and that it prevents references from modelling the concepts.

Ok.

I agree with the changes.

[rrahn]

looks good to me.

[rrahn]

is this done?

[h-2]

Yep