Core code independent of linear algebra implementation (#4)

* Initial work on abstract interface

* working code

* Move model definition to own file, make compute_stepsize more efficient

* Add tests

* Remove unused code

src/Cholesky/Cholesky.jl

@@ -0,0 +1,39 @@
+module Cholesky
+
+# export AbstractCholeskyFactor, SimpleDenseCholeskyFactor, SimpleSparseCholeskyFactor
+
+"""
+    cholesky!(A, d, M)
+    Form the matrix S = A*Diag(d)*A' and computes its Cholesky factorization.
+    The factorization overwrites F.
+"""
+function cholesky!(A::AbstractMatrix{Ta}, d::AbstractVector{Td}, F::Factorization{Ta}) where{Ta<:Real, Td<:Real}
+    warn("cholesky! must be implemented for $(typeof(A))")
+    return F
+end
+
+function cholesky!(A::Matrix{Ta}, d::AbstractVector{Td}, F::Base.LinAlg.Cholesky{Ta,Matrix{Ta}}) where{Ta<:Real, Td<:Real}
+    F = cholfact!(F, A*Diagonal(d)*A')
+    return F
+end
+
+function cholesky!(A::SparseMatrixCSC{Ta, Int64}, d::AbstractVector{Td}, F::Base.SparseArrays.CHOLMOD.Factor{Ta}) where{Ta<:Real, Td<:Real}
+    # update CHolesky factor
+    F = cholfact!(F, Symmetric(A*spdiagm(d)*A'))
+    return F
+end
+
+"""
+    solve!(y, F, b)
+    Solves the system F*y = b and overwrites y.
+"""
+function solve!(y::AbstractVector{T1}, F::Factorization{T2}, b::AbstractVector{T3}) where {T1<:Real, T2<:Real, T3<:Real}
+    warn("Implement solve! for concrete type implementations!")
+    return y
+end
+
+# include("denseCholesky.jl")
+# include("sparseCholesky.jl")
+
+
+end  # module

src/Tulip.jl

@@ -1,7 +1,13 @@
 module Tulip
 
+export Model, PrimalDualPoint, readmps
+
+# Cholesky module
+    include("Cholesky/Cholesky.jl")
+
 # package code goes here
-    include("./ipm.jl")
-    include("./readmps.jl")
+    include("model.jl")
+    include("ipm.jl")
+    include("readmps.jl")
 
 end # module