load_forecasts_repo(models = NULL) returns a data.frame where row order depends on locale

[Bisaloo]

You might decide that this is not your problem and the user is in charge of ensuring locale consistency across platforms and this would be a fair answer but I figured I could still open an issue just in case.
As illustrated in the reprex below, load_forecasts() output is locale-dependent because it is ordered according to the model column (when the models argument is absent from the load_forecasts() call) and this order change depending on the locale.
This causes unnecessary diffs (e.g., european-modelling-hubs/covid19-forecast-hub-europe_archive@be9803c) and potential confusion ("why am not getting the exact same result?") when contributors from different regions of the world / using different locales run the same piece of code.

I think an easy fix would be to change:

covidHubUtils/R/load_forecasts_repo.R

Line 50 in 1f97089

all_valid_models <- list.dirs(file_path, full.names = FALSE)

to

  all_valid_models <- sort(list.dirs(file_path, full.names = FALSE), method = "radix")

which always sorts according to a C locale and is thus locale-independent.

Reprex:

library(covidHubUtils)
library(dplyr)

withr::with_locale(c(LC_COLLATE = "fr_FR.UTF-8"), {
  d <- load_forecasts(
    source = "local_hub_repo",
    hub_repo_path = here::here(),
    hub = "ECDC"
  ) 
})

d %>% digest::digest()
#> [1] "e6329c2844d0ca6aa3e57c020b9509f8"
d %>%
  filter(model %in% c("MUNI-ARIMA", "MUNI_DMS-SEIAR")) %>%
  pull(model) %>%
  unique()
#> [1] "MUNI_DMS-SEIAR" "MUNI-ARIMA"


withr::with_locale(c(LC_COLLATE = "C"), {
  d <- load_forecasts(
    source = "local_hub_repo",
    hub_repo_path = here::here(),
    hub = "ECDC"
  ) 
})

d %>% digest::digest()
#> [1] "1a811419d52bb2d632b287ab88ffc7fd"

d %>%
  filter(model %in% c("MUNI-ARIMA", "MUNI_DMS-SEIAR")) %>%
  pull(model) %>%
  unique()
#> [1] "MUNI-ARIMA"     "MUNI_DMS-SEIAR"