We pack the bits representing the tableau in UInts and use bit-wise operations on them.

The library is set to support any sized UInt8/16/..., but we usually just use the default one.

There are a few tests that make sure that other sizes are still suported though (e.g. bitpacking and ecc_syndromes).

It would be useful to create the appropriate methods of reinterpret so that such tests are easier to write.

Here are some untested draft implementations (not a particularly good API design either):

    using QuantumClifford: Tableau
    reinterpret_frame(frame) = PauliFrame(reinterpret_stab(frame.frame), copy(frame.measurements))
    reinterpret_stab(s) = Stabilizer(Tableau(copy(phases(s)), nqubits(s), collect(reinterpret(UInt8, collect(s.tab.xzs)))[[1:1+(nqubits(s)-1)÷8;end÷2+1:end÷2+1+(nqubits(s)-1)÷8],:]))
    reinterpret_p(p) = PauliOperator(p.phase, nqubits(p), collect(reinterpret(UInt8, p.xz))[[1:1+(nqubits(p)-1)÷8;end÷2+1:end÷2+1+(nqubits(p)-1)÷8]])

This is also of great use when considering GPU code. (there should be "See also" entries to fastcolumn and fastrow and tocpu and togpu).