Specify extra repos in `pkg_install()`

[bwiernik]

With base install.packages(), it's possible to specify additional repos to search in addition to the ones listed in options("repos") with the repos argument:

install.packages('correlation', repos = 'https://easystats.r-universe.dev')

This is convenient, and I've found also easier to explain to my students and users of my packages. It would be nice to be able to do similarly with pkg_install() to add an extra repo to the list at function runtime:

pak::pkg_install('correlation', repos = 'https://easystats.r-universe.dev')

[gaborcsardi]

You can add extra repos with repo_add():

❯ pak::repo_add(easy = "https://easystats.r-universe.dev")
❯ pak::repo_get()
# A tibble: 6 x 5
  name        url                                  type   r_version bioc_version
  <chr>       <chr>                                <chr>  <chr>     <chr>
1 CRAN        https://cloud.r-project.org          cran   *         NA
2 easy        https://easystats.r-universe.dev     cranl… *         NA
3 BioCsoft    https://bioconductor.org/packages/3… bioc   4.1.0     3.13
4 BioCann     https://bioconductor.org/packages/3… bioc   4.1.0     3.13
5 BioCexp     https://bioconductor.org/packages/3… bioc   4.1.0     3.13
6 BioCworkfl… https://bioconductor.org/packages/3… bioc   4.1.0     3.13

❯ pak::meta_list("correlation")
# A tibble: 4 x 32
  package  version depends suggests   license imports   linkingto archs enhances
  <chr>    <chr>   <chr>   <chr>      <chr>   <chr>     <chr>     <chr> <chr>
1 correla… 0.6.1   R (>= … "BayesFac… GPL-3   "bayeste… NA        NA    NA
2 correla… 0.6.1   R (>= … "BayesFac… GPL-3   "bayeste… NA        NA    NA
3 correla… 0.6.2   R (>= … "BayesFac… GPL-3   "bayeste… NA        NA    NA
4 correla… 0.6.2   R (>= … "BayesFac… GPL-3   "bayeste… NA        NA    NA
# … with 23 more variables: os_type <chr>, priority <chr>,
#   license_restricts_use <chr>, license_is_foss <chr>, repodir <chr>,
#   platform <chr>, rversion <chr>, needscompilation <chr>, ref <chr>,
#   type <chr>, direct <lgl>, status <chr>, target <chr>, mirror <chr>,
#   sources <list>, filesize <int>, sha256 <chr>, sysreqs <chr>, built <chr>,
#   published <dttm>, deps <list>, md5sum <chr>, path <chr>

❯ pak::pkg_install('correlation')
✔ Updated metadata database: 1.11 MB in 3 files.
✔ Updating metadata database

→ Will install 5 packages.
→ Will download 3 CRAN packages (5.14 MB), cached: 2 (0 B).
+ bayestestR    0.10.0  ⬇ (3.29 MB)
+ correlation   0.6.2
+ datawizard    0.1.0   ⬇ (269.33 kB)
+ insight       0.14.2
+ parameters    0.14.0  ⬇ (1.57 MB)
ℹ Getting 3 pkgs (5.14 MB), 2 cached
✔ Got datawizard 0.1.0 (macos) (269.33 kB)
✔ Got parameters 0.14.0 (macos) (1.57 MB)
✔ Got insight 0.14.2 (macos) (1.74 MB)
✔ Got bayestestR 0.10.0 (macos) (3.29 MB)
✔ Got correlation 0.6.2 (macos) (3.69 MB)
✔ Downloaded 5 packages (10.57 MB)in 2.5s
✔ Installed datawizard 0.1.0  (115ms)
✔ Installed bayestestR 0.10.0  (234ms)
✔ Installed insight 0.14.2  (289ms)
✔ Installed parameters 0.14.0  (325ms)
✔ Installed correlation 0.6.2  (268ms)
✔ 1 + 4 pkgs | kept 0, updated 0, new 5 | downloaded 5 (10.57 MB) 14.1s

[bwiernik]

I'm aware of that. My point is that this is a more confusing operation for a one-off package installation:

pak::repo_add(easy = "https://easystats.r-universe.dev")
pak::pkg_install("correlation")

than:

pak::pkg_install("correlation", repos = "https://easystats.r-universe.dev")

Given the increasing prevalence of instructions like this (including in the pak readme):

install.packages("correlation", repos = "https://easystats.r-universe.dev")

it would help to eliminate friction if pak::pkg_install could be dropped in there without needing additional modifications:

pak::pkg_install("correlation", repos = "https://easystats.r-universe.dev")

A big part of my thinking there is that pak is becoming a much smoother approach to installation, and I am considering incorporating it into teaching to new R users as a unified way to install packages. Being able to say "if you're following instructions somewhere, you can replace either install.packages() or remotes::install_github() with pak::pkg_install() to take advantage of pak's headache-removing features" would be really useful.

[gaborcsardi]

I do not want to add many parameters to pkg_install(), because it clutters the UI. I would also need to add them to the 10-20 other pak functions that deal with repos, which seems like a lot of extra code and code paths, with only a small benefit. I think it is not too hard to set up the repo separately.

Also, I expect that people copy-paste installation instructions from the README, so I don't think it matters if it is one line or two.

Being able to say "if you're following instructions somewhere, you can replace either install.packages() or remotes::install_github() with pak::pkg_install() to take advantage of pak's headache-removing features" would be really useful.

That will only work in the simplest cases, because pkg_install() is not a drop-in replacements for these.

Btw.

install.packages(..., repos = "https://....r-universe.dev")

only works if the r-universe repo is self-contained, because it omits CRAN from your repos. So in reality you would need something like

install.packages(
  ..., 
  repos = c(
    runiverse = "https://....r-universe.dev", 
    CRAN = "https://cloud.r-project.org"
  )
)

This also gets more complicated if you have configured Bioconductor or other repos.