rocksdb and yas

[Martin-Lierschof]

rocksdb is really sometimes of a good serialization techique. Here yas can really shine in combination with rocksdb:slice.

rocksdb slice is functionally the same as yas::intrusive_buffer, so just to let you know and maybe put it in the documentation.

deserialization:

template <typename T>
T deserializeYas(const rocksdb::Slice& slice)
{
    T instance;
    if (slice->empty()) {
        return instance;
    }
    const yas::intrusive_buffer buf(slice.data(), slice.size());
    yas::load<yas::binary | yas::mem>(buf, instance);
    return instance;
}

const rocksdb::Slice sliceIn
MYSuperObject mysuperobject= rocksdb_tools::deserializeYas<std::vector<MYSuperObject >>(sliceIn);

serialization:

 template <typename T>
void serializeYas(const T& instance, yas::shared_buffer& buf)
{
      buf = yas::save<yas::binary | yas::mem>(instance);
}

yas::shared_buffer buf;
rocksdb_tools::serializeYas<MYSuperObject>(mysuperobject, buf);
db.Put(item.first, rocksdb::Slice(buf.data.get(), buf.size));

Obviously we could improve this behaviour by directly adding support for rocksdb::slice to to be used as input and output buffer.
Therefore you would need to find a way to manage the shared_buffer inside the scope of a rocksdb::slice.

[niXman]

hi,

yeah, it looks like std::span<char>: https://en.cppreference.com/w/cpp/container/span

but I don't see any point in adding support for this type because it's a very rare case...
I have a plan to add support for std::span to YAS, and I think it would be a good idea for the RockDB developers to also switch to using std::span instead of rocksdb::Slice.