Invalid LAPACK call (via eigvals) in welch with large nsegments

[Andy Nowacki]

welch ends up erroring in a LAPACK call when trying to use large values for nsegments:

julia> welch(randn(100), 50)
 ** On entry to DSTEMRM parameter number  8 had an illegal value
ERROR: ArgumentError: invalid argument #8 to LAPACK call
Stacktrace:
 [1] chklapackerror at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.5/LinearAlgebra/src/lapack.jl:36 [inlined]
 [2] stegr!(::Char, ::Char, ::Array{Float64,1}, ::Array{Float64,1}, ::Float64, ::Float64, ::Int64, ::Int64) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.5/LinearAlgebra/src/lapack.jl:3819
 [3] eigvals! at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.5/LinearAlgebra/src/tridiag.jl:295 [inlined]
 [4] eigvals(::LinearAlgebra.SymTridiagonal{Float64,Array{Float64,1}}, ::UnitRange{Int64}) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.5/LinearAlgebra/src/tridiag.jl:297
 [5] dpss_tapers(::Int64, ::Float64, ::Int64, ::Symbol) at /Users/nowacki/.julia/dev/Multitaper/src/Univariate.jl:48
 [6] dpss_tapers at /Users/nowacki/.julia/dev/Multitaper/src/Univariate.jl:45 [inlined]
 [7] welch(::Array{Float64,1}, ::Int64, ::Float64; NW::Float64, K::Int64, dt::Float64, ctr::Bool, pad::Float64, dpVec::Nothing, egval::Nothing, guts::Bool, a_weight::Bool, Ftest::Bool, jk::Bool, Tsq::Nothing, alph::Float64) at /Users/nowacki/.julia/dev/Multitaper/src/Univariate.jl:608
 [8] welch at /Users/nowacki/.julia/dev/Multitaper/src/Univariate.jl:599 [inlined] (repeats 2 times)
 [9] top-level scope at REPL[17]:1

I don’t know if it feasible to check for inappropriate nsegments values, but throwing an error there may be more user-friendly than the current behaviour.

[Note that this is a bug report, not part of the JOSS review.]

[Chris Geoga]

Thanks for pointing this out. We’ve added an assertion check in 8ca7898.

[Andy Nowacki]

That’s great.

In my experience it is more usual in idiomatic Julia to throw an ArgumentError for invalid user input than an AssertionError. The latter, which is what the @assert macro does, is more for cases which should never be true by design or in prototype code.

[Chris Geoga]

Oof, I waffled about which one to use. Thanks for the pointer. I’ve corrected it to ArgumentError in the most recent commit.