error reported by CRAN check

The following error appears on the CRAN package check results for one of the debian architectures:

── Failure ('test_transform.R:16:9'): Function pullback() behaves itself ───────
     issmall(discrepancy) is not TRUE

     `actual`: FALSE
     `expected`: TRUE
     list(index = c(1, 3, 5, 5), value = c(-0.170070938205061, -0.800587236566934))
     c(1.46372997954945, 0.4755026057719, 1.18200674579991, 0.197517801699613, -0.276373658487993, 1.1515112794683, -0.187898172031518, 0.924765696304977, -0.232144224280786, 0.96302634681183, 1.30899841041773, 0.621688176410363, -0.324975905380696, -0.0389848423003657, -0.940345853991544, -0.948041906536554, -0.408115650086504, 0.630512116576583, -0.0134038963463815, -0.561589637973832, 1.53093308774001, 0.922058624536413, 0.418724053361834, 0.810188877253954, 2.40135084607294)
     Backtrace:
     ▆
     1. └─stokes (local) foo(o, M) at test_transform.R:16:8
     2. └─testthat::expect_true(issmall(discrepancy), info = list(x, M)) at test_transform.R:10:8

  [snip]

For ease of reference, here is test_transform.R:

## Some tests of pullback()

options(warn=999)
test_that("Function pullback() behaves itself", {
    expect_true(TRUE)

    foo <- function(x,M){  # checks that pullback(pullback(x,M),solve(M)) == x
        xt <- pullback(pullback(x,M),solve(M))
        discrepancy <- x |> pullback(M) |> pullback(solve(M)) - x
        expect_true(issmall(discrepancy),info=list(x,M))
    } # foo() closes

    for(i in 1:3){
        o <-  rform(terms=3,k=2,n=5,coeffs=rnorm(3))
        M <- matrix(rnorm(25),5,5)
        foo(o,M)
        foo(o*0,M)
    }
})

The information given by testthat is somewhat opaque, but we can generate the error from the dput() as follows:

library("stokes")
x <- kform(spray(matrix(c(1, 3, 5, 5),2,2),c(-0.170070938205061, -0.800587236566934)))
M <- matrix(
c(1.46372997954945, 0.4755026057719, 1.18200674579991, 0.197517801699613, -0.276373658487993, 1.1515112794683, -0.187898172031518, 0.924765696304977, -0.232144224280786, 0.96302634681183, 1.30899841041773, 0.621688176410363, -0.324975905380696, -0.0389848423003657, -0.940345853991544, -0.948041906536554, -0.408115650086504, 0.630512116576583, -0.0134038963463815, -0.561589637973832, 1.53093308774001, 0.922058624536413, 0.418724053361834, 0.810188877253954, 2.40135084607294),5,5)
discrepancy <- x |> pullback(M) |> pullback(solve(M)) - x
discrepancy
#> An alternating linear map from V^2 to R with V=R^5:
#>             val
#>  1 3  =   0e+00
#>  3 4  =   1e-07
#>  4 5  =   0e+00
#>  2 5  =   0e+00
#>  3 5  =   0e+00
#>  2 4  =  -1e-07
#>  1 5  =   0e+00
#>  1 4  =  -1e-07
#>  1 2  =   0e+00
#>  2 3  =   0e+00
coeffs(discrepancy)
#> A disord object with hash 5e480a505fb8e76aca638d263498e13b4bb737d9 and elements
#>  [1]  1.138505e-08  5.612856e-08  8.429197e-09  1.226908e-08  2.023534e-09
#>  [6] -9.264841e-08 -2.081617e-09 -6.275604e-08 -2.283195e-08 -3.348759e-08
#> (in some order)
issmall(discrepancy)
#> [1] FALSE

(the value of x is passed as variable o by foo()). The problem is caused by the small determinant of matrix M:

 det(M)
# [1] 0.0001320315

although tbh this is not particularly small, but presumably M is close to some problematic singular value.

RobinHankin / stokes

error reported by CRAN check #81