Regression in `===` (caused by #43658?)

[KristofferC]

Running the following:

using TensorKit, TensorKitManifolds, StableRNGs
V = ℂ^4
T = Float64
rng = StableRNG(0)
W, = leftorth(TensorMap((x...)-> randn(rng, x...), T, V*V*V, V*V); alg = Polar())
X = TensorMap((x...)-> randn(rng, x...), T, space(W))
Y = TensorMap((x...)-> randn(rng, x...), T, space(W))
Δ = Unitary.project(X, W)
Θ = Unitary.project(Y, W)

Δ === Θ

we get falseon 1.7, while on 1.8 this returns true

Here is a proof they are not ===:

julia> Θ.A.data |> sum
1.1102230246251565e-15

julia> Δ.A.data |> sum
-4.6074255521944e-15

Reverting #43658 makes it go away so I guess the bug is from there.

cc @NHDaly

[vtjnash]

julia/src/builtins.c

Line 110 in 62e0729

return 1;

😱

[NHDaly]

julia/src/builtins.c

Line 110 in 62e0729

return 1;

😱

😮
Interestingly, I didn't change that line in my PR #43658. Was #43658 somehow not the original cause of the bug, then? It maybe just exposed this existing bug more often due to the changed order we check things in?

[Moelf]

Yeah I think if that happens last it's not a problem