in_parallel() causes R-CMD-check error on Windows

[padpadpadpad]

I am parallelising some code for an R package I maintain. I want to resubmit this to CRAN, but it fails the GitHub Actions R-CMD-check, but only for Windows latest release.

All the other platforms pass. The code that fails is the in_parallel() code I think, I get the error:

ℹ In argument: `mods = map(...)`.
Caused by error in `map()`:
ℹ In index: 1.
Caused by error in `library()`:
! there is no package called 'rTPC' 

But I cannot work out why. Reproducible code is below. Is there something I am missing? quickfit_tpc_multi() is a function of rTPC, I have also tried rTPC::quickfit_tpc_multi() and the same problem persists.

# load in packages
library(rTPC)
library(nls.multstart)
library(broom)
library(tidyverse)
library(mirai)

# load in data
data('chlorella_tpc')
d <- chlorella_tpc
head(d)

# start timer
parallel_start <- Sys.time()

# start parallel session
daemons(3)

# fit models as before, but in parallel
d_mods_para <- d %>%
  nest(data = c(temp, rate)) %>%
  mutate(
    mods = map(
      data,
      in_parallel(
        \(x) {
          library(rTPC)
          quickfit_tpc_multi(
            data = x,
            c(
              "briereextended_2021",
              "gaussianmodified_2006",
              "briere1simplified_1999",
              "tomlinsonphillips_2015",
              'lactin2_1995'
            ),
            "temp",
            "rate",
            iter = list(150, 150, c(4, 4, 4), 150, 200),
            lhstype = 'random'
          )
        }
      ),
      .progress = TRUE
    )
  ) %>%
  tidyr::unnest(mods)

# close parallel session
daemons(0)

# end time of parallel
parallel_end <- Sys.time()

[shikokuchuo]

First, when producing vignettes, please follow the guidance to not use more than 2 cores on CRAN - note that this applies to vignettes as well as tests. It's often enough to demonstrate how to run things in parallel without actually doing so.

If you actually want to show timings, then pre-computing the vignettes can be a good option so that they don't run on CRAN. For how to do so, you may refer to: https://ropensci.org/blog/2019/12/08/precompute-vignettes/

As for the failure, I did try to re-produce on a Windows machine, and it seems to be elsewhere in the code. Even if I add eval=FALSE to your entire parallel block, I still get an error. So you should probably look to simplify the example as far as possible.