Abstract type for matrices wrapping a single vector

[jishnub]

Defining this abstract type would mean that several functions need to only be extended for the abstract type instead of for every concrete subtype. Also, future types subtyping the abstract type get these methods for free.

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.26 🎉

Comparison is base (7558d4a) 94.72% compared to head (eb6c504) 94.98%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #98      +/-   ##
==========================================
+ Coverage   94.72%   94.98%   +0.26%     
==========================================
  Files           7        7              
  Lines         796      798       +2     
==========================================
+ Hits          754      758       +4     
+ Misses         42       40       -2     

Impacted Files	Coverage Δ
src/ToeplitzMatrices.jl	92.30% <ø> (-7.70%)	⬇️
src/hankel.jl	96.15% <100.00%> (+2.48%)	⬆️
src/special.jl	99.29% <100.00%> (+0.74%)	⬆️
src/toeplitz.jl	97.53% <100.00%> (-0.04%)	⬇️

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[putianyi889]

Could this idea be extended to a new package, say LinearStructuredMatrices (or maybe just add to ArrayInterfaces)? They have a parent where they linearly depend on.

[jishnub]

Does that help in general? In this case, we know the structure of the matrices and can forward various functions to the parent.

[putianyi889]

Does that help in general? In this case, we know the structure of the matrices and can forward various functions to the parent.

for example, Diagonal depending on one vector, Bidiagonal depending on a tuple of vectors, CircularArray depending on a matrix, Adjoint depending on a vector or matrix.

"linear" could be too general for most cases. A subtype would be that the values of parent are exactly projected to different locations.

[jishnub]

This might be worth discussing in ArrayLayouts, as, IIUC, you're suggesting field traits aside from the existing structure traits. It might be overkill for this package at present, but if such a trait is added elsewhere, we may use it here