Missing dependency libgsl.so

[trevorcampbell]

Hi there!

I've been playing around with tidyclust, and it looks really neat. I think I found a missing dependency of libgsl.so -- it's a bit of an unlikely case, because most machines will already have it, but I'm running things in a pared down Docker container which didn't have it originally.

Reprex

If I build this Dockerfile (there are probably more packages than necessary to reproduce there, but this was the Dockerfile for the image I was working with):

FROM rocker/verse:4.3.1

RUN apt-get update --fix-missing
RUN Rscript -e "update.packages(ask = FALSE)"
RUN install2.r --error magick \
                       cowplot \
                       kableExtra \
                       plotly \
                       tidymodels \
                       tidyclust \
                       kknn \
                       ggpubr \
                       ggforce \
                       themis \
                       egg \
                       fontawesome \
                       xfun \
                       tinytex \
                       reticulate \
                       rsvg

and run the following code

library(tidyverse)
library(tidymodels)
library(tidyclust)

penguins <- read_csv("penguins.csv")
spec <- k_means(num_clusters = tune()) |> set_engine("stats")
rec <- recipe(~ ., data=penguins) |> step_center(all_predictors()) |> step_scale(all_predictors())
boots <- bootstraps(penguins, times = 10)
workflow() |>
    add_recipe(rec) |>
    add_model(spec) |>
    tune_cluster(resamples = boots)

where penguins.csv contains

bill_length_mm,flipper_length_mm
39.2,196
36.5,182
34.5,187
36.7,187
38.1,181
39.2,190
36,195
37.8,193
46.5,213
46.1,215
47.8,215
45,220
49.1,212
43.3,208
46,195
46.7,195
52.2,197
46.8,189

I get the following error:

x Bootstrap01: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap02: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap03: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap04: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap05: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap06: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap07: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap08: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap09: internal: Error in `value[[3L]]()`:
! In me...
x Bootstrap10: internal: Error in `value[[3L]]()`:
! In me...

Investigating it a bit, I see that libgsl.so appears to be missing. I would expect tidyclust to install that as a dependency when it's installed.

f$.notes[[8]]$note
[1] "\033[1m\033[33mError\033[39m in `value[[3L]]()`:\033[22m\n\033[33m!\033[39m In metric: `sse_within_total`\nunable to load shared object '/usr/local/lib/R/site-library/RcppZiggurat/libs/RcppZiggurat.so':\n  libgsl.so.27: cannot open shared object file: No such file or directory"

If I then run

apt install libgsl27

in the container and then re-run the above R code, everything seems to work OK.

Versions

R 4.3.1
tidyverse 2.0.0
tidymodels 1.1.0
tidyclust 0.1.2
tune 1.1.1
parsnip 1.1.0
workflows 1.1.3

[EmilHvitfeldt]

Hello @trevorcampbell Thank you for the report, and the fine detective work!

When looking at the error message we see '/usr/local/lib/R/site-library/RcppZiggurat/libs/RcppZiggurat.so' leading me to suspect that this is an bug in {RcppZiggurat} rather than {tidyclust}.

pak::pkg_deps_explain("tidyclust", "RcppZiggurat")
#> ℹ Loading metadata database
#> ✔ Loading metadata database ... done
#> 
#> tidyclust -> Rfast -> RcppZiggurat

I would suggest you file an issue over in that repository https://github.com/eddelbuettel/rcppziggurat

[eddelbuettel]

leading me to suspect that this is an bug in RcppZiggurat

I take the other side in eddelbuettel/rcppziggurat#19

[eddelbuettel]

What must have happened here is that a installation notice and error, likely from RcppGSL which has the 'hard' GSL dependendy, was ignored. What we have here is the usual issue of an R installation being unable to guarantee a system dependency. The minimally verifiable complete example given is based on rocker/r-verse which uses PPM/RSPM to install, and that can no 'real' connection to the system. If we use rocker/r2u, which does, it all works.

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.