gausshermite(21) returns error

[the-noble-argon]

gausshermite(20) works, gausshermite(22) works, it appears 21 is the unlucky number:

ERROR: BoundsError: attempt to access 10-element Vector{Float64} at index [11]
Stacktrace:
[1] setindex!
@ .\array.jl:839 [inlined]
[2] setindex!
@ .\array.jl:853 [inlined]
[3] hermite_initialguess(n::Int64)
@ FastGaussQuadrature C:\Users.julia\packages\FastGaussQuadrature\WqyQC\src\gausshermite.jl:271
[4] hermite_rec(n::Int64)
@ FastGaussQuadrature C:\Users.julia\packages\FastGaussQuadrature\WqyQC\src\gausshermite.jl:86
[5] unweightedgausshermite(n::Int64)
@ FastGaussQuadrature C:\Users.julia\packages\FastGaussQuadrature\WqyQC\src\gausshermite.jl:41
[6] gausshermite(n::Int64)
@ FastGaussQuadrature C:\Users.julia\packages\FastGaussQuadrature\WqyQC\src\gausshermite.jl:23
[7] top-level scope
@ REPL[7]:1

[hyrodium]

This bug was introduced in release v0.4.8.

julia> using FastGaussQuadrature  # On v0.4.8

julia> gausshermite(21)
ERROR: BoundsError: attempt to access 10-element Vector{Float64} at index [11]
Stacktrace:
 [1] setindex!
   @ ./array.jl:843 [inlined]
 [2] setindex!
   @ ./array.jl:857 [inlined]
 [3] hermite_initialguess(n::Int64)
   @ FastGaussQuadrature ~/.julia/dev/FastGaussQuadrature/src/gausshermite.jl:271
 [4] hermite_rec(n::Int64)
   @ FastGaussQuadrature ~/.julia/dev/FastGaussQuadrature/src/gausshermite.jl:86
 [5] unweightedgausshermite(n::Int64)
   @ FastGaussQuadrature ~/.julia/dev/FastGaussQuadrature/src/gausshermite.jl:41
 [6] gausshermite(n::Int64)
   @ FastGaussQuadrature ~/.julia/dev/FastGaussQuadrature/src/gausshermite.jl:23
 [7] top-level scope
   @ REPL[2]:1

julia> using FastGaussQuadrature  # On v0.4.7

julia> gausshermite(21)
([-5.550351873264678, -4.773992343411219, -4.1219955474918395, -3.5319728771376777, -2.979991207704598, -2.453552124512838, -1.9449629491862535, -1.448934250650732, -0.9614996344183689, -0.4794507070791075  …  0.4794507070791075, 0.9614996344183689, 1.448934250650732, 1.9449629491862535, 2.453552124512838, 2.979991207704598, 3.5319728771376777, 4.1219955474918395, 4.773992343411219, 5.550351873264678], [3.7203650688183795e-14, 8.818611242049893e-11, 2.5712301800593306e-8, 2.1718848980566762e-6, 7.478398867310059e-5, 0.0012549820417264083, 0.011414065837434388, 0.060179646658912324, 0.1921203240669979, 0.381669073613502  …  0.381669073613502, 0.1921203240669979, 0.060179646658912324, 0.011414065837434388, 0.0012549820417264083, 7.478398867310059e-5, 2.1718848980566762e-6, 2.5712301800593306e-8, 8.818611242049893e-11, 3.7203650688183795e-14])

[hyrodium]

Thanks for the bug report! I made a PR to fix this bug.

[bvdmitri]

Thanks for the fast fix! Could you please be so kind to tag a new release with new fix?

This bug breaks our package for Bayesian inference on precompilation stage, because apparently we used 21 number of points as a default value 😅😅.

https://github.com/biaslab/ReactiveMP.jl/blob/77d9fae7d9466d92e63731d89ac086c2e15905b4/src/nodes/gcv.jl#L17

[dlfivefifty]

done!