Fix IterativeSolvers.jl for Julia 0.7 (#162)

* Escape backslash

* Fix Bidiagonal deprecation

* Comment out the assertion that breaks on 0.7

* Test whether convergence history objects are returned when log = true

* Check uplo on .7 and isupper on .6

Author: haampie

src/lsmr.jl

@@ -12,7 +12,7 @@ lsmr(A, b; kwargs...) = lsmr!(zerox(A, b), A, b; kwargs...)
 """
     lsmr!(x, A, b; kwargs...) -> x, [history]
 
-Minimizes ``\|Ax - b\|^2 + \|λx\|^2`` in the Euclidean norm. If multiple solutions
+Minimizes ``\\|Ax - b\\|^2 + \\|λx\\|^2`` in the Euclidean norm. If multiple solutions
 exists the minimum norm solution is returned.
 
 The method is based on the Golub-Kahan bidiagonalization process. It is 

src/lsqr.jl

@@ -10,7 +10,7 @@ lsqr(A, b; kwargs...) = lsqr!(zerox(A, b), A, b; kwargs...)
 """
     lsqr!(x, A, b; kwargs...) -> x, [history]
 
-Minimizes ``\|Ax - b\|^2 + \|damp*x\|^2`` in the Euclidean norm. If multiple solutions
+Minimizes ``\\|Ax - b\\|^2 + \\|damp*x\\|^2`` in the Euclidean norm. If multiple solutions
 exists returns the minimal norm solution.
 
 The method is based on the Golub-Kahan bidiagonalization process.

src/svdl.jl

@@ -356,9 +356,8 @@ function build(log::ConvergenceHistory, A, q::AbstractVector{T}, k::Int) where {
     p = A*q
     α = convert(Tr, norm(p))
     @blas! p *= inv(α)
-    extend!(log, A, PartialFactorization(
-        reshape(p, m, 1), reshape(q, n, 1), Bidiagonal([α], Tr[], true), β
-        ), k)
+    bidiag = Bidiagonal([α], Tr[], @static VERSION < v"0.7.0-DEV.884" ? true : :U)
+    extend!(log, A, PartialFactorization(reshape(p, m, 1), reshape(q, n, 1), bidiag, β), k)
 end
 
 
@@ -445,7 +444,9 @@ function harmonicrestart!(A, L::PartialFactorization{T,Tr},
     #Compute scaled residual from the harmonic Ritz problem
     r0 = zeros(Tr, m)
     r0[end] = 1
-    isa(L.B, Bidiagonal) && @assert L.B.isupper
+    if isa(L.B, Bidiagonal)
+        @assert @static VERSION < v"0.7.0-DEV.884" ? L.B.isupper : L.B.uplo == 'U'
+    end
     r = try
         #(L.B\r0)
         A_ldiv_B!(L.B, r0)

test/bicgstabl.jl

@@ -16,6 +16,7 @@ n = 20
     @testset "BiCGStab($l)" for l = (2, 4)
         # Solve without preconditioner
         x1, his1 = bicgstabl(A, b, l, max_mv_products = 100, log = true, tol = tol)
+        @test isa(his1, ConvergenceHistory)
         @test norm(A * x1 - b) / norm(b) ≤ tol
 
         # Do an exact LU decomp of a nearby matrix

test/cg.jl

@@ -18,6 +18,7 @@ srand(1234321)
         tol = √eps(real(T))
 
         x,ch = cg(A, b; tol=tol, maxiter=2n, log=true)
+        @test isa(ch, ConvergenceHistory)
         @test norm(A*x - b) / norm(b) ≤ tol
         @test ch.isconverged
 

test/chebyshev.jl

@@ -26,6 +26,7 @@ srand(1234321)
     tol = √(eps(real(T)))
 
     x, history = chebyshev(A, b, λ_min, λ_max, tol=tol, maxiter=10n, log=true)
+    @test isa(history, ConvergenceHistory)
     @test history.isconverged
     @test norm(A * x - b) / norm(b) ≤ tol
 

test/gmres.jl

@@ -17,6 +17,7 @@ n = 10
 
     # Test optimality condition: residual should be non-increasing
     x, history = gmres(A, b, log = true, restart = 3, maxiter = 10, tol = tol);
+    @test isa(history, ConvergenceHistory)
     @test all(diff(history[:resnorm]) .<= 0.0)
 
     # Left exact preconditioner

test/idrs.jl

@@ -14,6 +14,7 @@ srand(1234567)
 
     @testset "Without residual smoothing" begin
         x, history = idrs(A, b, log=true)
+        @test isa(history, ConvergenceHistory)
         @test history.isconverged
         @test norm(A * x - b) / norm(b) ≤ tol
     end

test/lsmr.jl

@@ -101,7 +101,8 @@ end
     @testset "Small dense matrix" for T = (Float32, Float64)
         A = rand(T, 10, 5)
         b = rand(T, 10)
-        x = lsmr(A, b)
+        x, history = lsmr(A, b, log = true)
+        @test isa(history, ConvergenceHistory)
         @test norm(x - A\b) ≤ √eps(T)
     end
 

test/lsqr.jl

@@ -10,6 +10,7 @@ srand(1234321)
     A = rand(T, 10, 5)
     b = rand(T, 10)
     x, history = lsqr(A, b, log = true)
+    @test isa(history, ConvergenceHistory)
     @test norm(x - A\b) ≤ √eps(T)
     @test history.isconverged
     @test last(history[:resnorm]) ≈ norm(b - A * x) atol=√eps(T)