[Search] Shapes are the wrong way around

[MitraDarja]

I think the kmer hash view does not work with views, which have shapes in the form 1101 or 1011.

#include <seqan3/alphabet/nucleotide/dna4.hpp>
#include <seqan3/core/debug_stream.hpp>
#include <seqan3/search/views/kmer_hash.hpp>

int main()
{
    using namespace seqan3::literals;

    std::vector<seqan3::dna4> const text{"ACGTC"_dna4};
    auto const shape1011 = seqan3::shape{seqan3::bin_literal{0b1011}};
    auto const shape1101 = seqan3::shape{seqan3::bin_literal{0b1101}};

    seqan3::debug_stream << "shape1011\nExpected: [11,29]\nActual  : "
                         << (text | seqan3::views::kmer_hash(shape1011)) << "\n\n";

    seqan3::debug_stream << "shape1101\nExpected: [7,25]\nActual  : "
                         << (text | seqan3::views::kmer_hash(shape1101)) << '\n';
}

shape1011
Expected: [11,29]
Actual  : [7,25]

shape1101
Expected: [7,25]
Actual  : [11,29]

My assumption is that the shift does not work correctly with these shapes, the only shapes we tested so far are shapes of the sort 1(any number of zeros)1.

[eseiler]

The values seem correct to me?

1011
ACGTC
 1011

------------------------------

16 4  1           16 4  1 
A  G  T           C  T  C
0  2  3           1  3  1

------------------------------

0 + 8 + 3 = 11    16 + 12 + 1 = 29

1101
ACGTC
 1101

------------------------------

16 4  1           16 4  1 
A  C  T           C  G  T
0  1  3           1  2  1

------------------------------

0 + 4 + 3 = 7     16 + 8 + 1 = 25

[MitraDarja]

@eseiler

Maybe, I misunderstand the shape, isn't 0b1101 1101 (so your second example) and we get the result of the first and vice visa?

[eseiler]

Ok, I get it now that the snippet is correct :)

The problem is that this inherits from the dynamic_bitset (which is just a constexpr-capable version of std::bitset, and implements the same API).

This means that

seqan3::dynamic_bitset t1{0b1011'1000'1111};
seqan3::debug_stream << t1 << '\n'; // 1011'1000'1111
seqan3::debug_stream << t1[1] << '\n'; // 1

I.e., it prints left to right, but the indices go right to left (unless this is the actual bug, would need to check std::bitset).

So, we could just call this->flip() in the shape-ctor https://github.com/seqan/seqan3/blob/master/include/seqan3/search/kmer_index/shape.hpp#L105
but then the shape would be printed the wrong way around, except when we overload.

[MitraDarja]

Sorry, for the wrong snippet, I got myself confused. 😅

Maybe, we mention this in the documentation?

[eseiler]

I reopen this because that shapes are "the wrong way around" might be a problem :)

[SGSSGene]

Maybe just adding some documentation would be enough? #2981

[eseiler]

Maybe just adding some documentation would be enough? #2981

Yes, at the very least 👍 . We can talk in the meeting about it.
Either we find that this is enough, or we want some special behaviour that makes it behave like you would expect :D

Edit:

How do shapes work in SeqAn2?

[smehringer]

Might be of interest when discussing this seqan/product_backlog#388