Regression with [email protected]

[sathvikbhagavan]

An issue was caught in SciML/DataInterpolations.jl#178 where tests started failing with [email protected]

I have made a MWE to illustrate the issue (using a simple LinearInterpolation from DataInterpolations):

u = collect(0:10)
t = collect(0:10)
A = LinearInterpolation(u,t)
qint, err = quadgk(A, 0.0, 10.0)

It is a straight line with slope 1. So the integral should be 50.0.

with [email protected]

julia> qint, err = quadgk(A, 0.0, 10.0)
(50.0, 0.0)

with [email protected]

julia> qint, err = quadgk(A, 0.0, 10.0)
(51.76911234248546, 4.274582013885939)

[stevengj]

cc @lxvm

[stevengj]

Seems to be a dispatch problem, since passing the argument as a function works:

julia> quadgk(A, 0.0, 10.0)
(51.76911234248546, 4.274582013885939)

julia> quadgk(t -> A(t), 0.0, 10.0)
(50.0, 0.0)

[stevengj]

Okay, the problem is that LinearInterpolation is a subtype of AbstractVector, so it is calling the wrong method of QuadGK.evalrule.

[lxvm]

Right, I can rename evalrule(::AbstractVector,...) to batchevalrule to fix the issue

[stevengj]

Should be fixed by #90. Will merge and tag when green.

[lxvm]

Thanks!