`upgrade=FALSE` ignores system-wide installed packages

[assaron]

Hello,

Since devtools::install_dev_deps is going to be deprecated in favor of pak::local_install_dev_deps I tried it and stumbled upon the problem that pak ignores any already system-wide installed packages (in my case installed by r2u in a docker environment). I understand from the documentation it's partially by design and follows from lib = .libPaths()[1] in the arguments.

Is it really by design? It's very confusing to me at the moment. Is there a way to account for all the packages and install only the packages that are really absent?

Thanks,
Alexy

[BjarkeHautop]

From r-lib/pkgdepends@634661a, which was added in pak in a041d39, you can pass a vector of libraries now. See reprex below (same idea for other pak functions):

lib  <- tempfile()
lib2 <- tempfile()

dir.create(lib)
dir.create(lib2)

pak::pak("ggplot2", lib = lib)
#> 
#> ✔ Updated metadata database: 4.21 MB in 6 files.
#> 
#> ℹ Updating metadata database
#> ✔ Updating metadata database ... done
#> 
#> 
#> → Will install 16 packages.
#> → Will download 16 packages with unknown size.
#> + cli            3.6.5 [dl]
#> + farver         2.1.2 [dl]
#> + ggplot2        4.0.2 [dl]
#> + glue           1.8.0 [dl]
#> + gtable         0.3.6 [dl]
#> + isoband        0.3.0 [dl]
#> + labeling       0.4.3 [dl]
#> + lifecycle      1.0.5 [dl]
#> + R6             2.6.1 [dl]
#> + RColorBrewer   1.1-3 [dl]
#> + rlang          1.1.7 [dl]
#> + S7             0.2.1 [dl]
#> + scales         1.4.0 [dl]
#> + vctrs          0.7.2 [dl]
#> + viridisLite    0.4.3 [dl]
#> + withr          3.0.2 [dl]
#> ℹ Getting 16 pkgs with unknown sizes
#> ✔ Cached copy of cli 3.6.5 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of farver 2.1.2 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of ggplot2 4.0.2 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of glue 1.8.0 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of gtable 0.3.6 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of isoband 0.3.0 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of labeling 0.4.3 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of R6 2.6.1 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of lifecycle 1.0.5 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of RColorBrewer 1.1-3 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of rlang 1.1.7 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of scales 1.4.0 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of S7 0.2.1 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of vctrs 0.7.2 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of viridisLite 0.4.3 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of withr 3.0.2 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Installed cli 3.6.5  (184ms)
#> ✔ Installed farver 2.1.2  (210ms)
#> ✔ Installed ggplot2 4.0.2  (225ms)
#> ✔ Installed glue 1.8.0  (232ms)
#> ✔ Installed gtable 0.3.6  (238ms)
#> ✔ Installed isoband 0.3.0  (256ms)
#> ✔ Installed labeling 0.4.3  (257ms)
#> ✔ Installed lifecycle 1.0.5  (259ms)
#> ✔ Installed R6 2.6.1  (261ms)
#> ✔ Installed RColorBrewer 1.1-3  (263ms)
#> ✔ Installed rlang 1.1.7  (285ms)
#> ✔ Installed S7 0.2.1  (290ms)
#> ✔ Installed scales 1.4.0  (293ms)
#> ✔ Installed vctrs 0.7.2  (295ms)
#> ✔ Installed viridisLite 0.4.3  (299ms)
#> ✔ Installed withr 3.0.2  (355ms)
#> ✔ 1 pkg + 15 deps: added 16 [8.6s]
pak::pak("ggplot2", lib = c(lib2, lib))
#> 
#> ℹ No downloads are needed
#> ✔ 1 pkg + 15 deps: kept 16 [299ms]
pak::pak("ggplot2", lib = lib2)
#> 
#> → Will install 16 packages.
#> → Will download 16 packages with unknown size.
#> + cli            3.6.5 [dl]
#> + farver         2.1.2 [dl]
#> + ggplot2        4.0.2 [dl]
#> + glue           1.8.0 [dl]
#> + gtable         0.3.6 [dl]
#> + isoband        0.3.0 [dl]
#> + labeling       0.4.3 [dl]
#> + lifecycle      1.0.5 [dl]
#> + R6             2.6.1 [dl]
#> + RColorBrewer   1.1-3 [dl]
#> + rlang          1.1.7 [dl]
#> + S7             0.2.1 [dl]
#> + scales         1.4.0 [dl]
#> + vctrs          0.7.2 [dl]
#> + viridisLite    0.4.3 [dl]
#> + withr          3.0.2 [dl]
#> ℹ Getting 16 pkgs with unknown sizes
#> ✔ Cached copy of cli 3.6.5 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of farver 2.1.2 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of ggplot2 4.0.2 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of glue 1.8.0 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of gtable 0.3.6 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of isoband 0.3.0 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of labeling 0.4.3 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of S7 0.2.1 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of lifecycle 1.0.5 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of R6 2.6.1 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of RColorBrewer 1.1-3 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of rlang 1.1.7 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of viridisLite 0.4.3 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of scales 1.4.0 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of vctrs 0.7.2 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Cached copy of withr 3.0.2 (x86_64-pc-linux-gnu-linuxmint-22.3) is the latest build
#> ✔ Installed cli 3.6.5  (181ms)
#> ✔ Installed farver 2.1.2  (206ms)
#> ✔ Installed ggplot2 4.0.2  (219ms)
#> ✔ Installed glue 1.8.0  (225ms)
#> ✔ Installed gtable 0.3.6  (230ms)
#> ✔ Installed isoband 0.3.0  (236ms)
#> ✔ Installed labeling 0.4.3  (254ms)
#> ✔ Installed lifecycle 1.0.5  (254ms)
#> ✔ Installed R6 2.6.1  (255ms)
#> ✔ Installed RColorBrewer 1.1-3  (257ms)
#> ✔ Installed rlang 1.1.7  (274ms)
#> ✔ Installed S7 0.2.1  (277ms)
#> ✔ Installed scales 1.4.0  (278ms)
#> ✔ Installed vctrs 0.7.2  (280ms)
#> ✔ Installed viridisLite 0.4.3  (282ms)
#> ✔ Installed withr 3.0.2  (302ms)
#> ✔ 1 pkg + 15 deps: added 16 [2s]

^{Created on 2026-03-24 with reprex v2.1.1}

But otherwise, you can also pass dependencies = FALSE to pak::local_install_deps() (assuming the dependencies are installed in another library) (related to r-lib/devtools#2683).

[assaron]

Oh, great, thank you! Is there any particular reason why .libPaths()[1] is the default value and not the whole .libPaths()?

On a less related note: how does it relate to the remotes package? When stumbled upon the original problem, I switched to remotes::install_dependencies() in my workflow, following the current devtools implementation. Will it be deprecated too at some point?

[BjarkeHautop]

Is there any particular reason why .libPaths()[1] is the default value and not the whole .libPaths()?

To avoid potential ambiguities/problems with the same package installed in different libraries, which could cause packages to be unable to load, I believe. See https://pak.r-lib.org/reference/pak_solver.html and imagine the package cli was installed in two libraries, which could cause you to be unable to load pkgcache via library() if cli@message was in the first entry of lib.
By installing everything into a single library (the default), all packages in that library are consistent. This approach is similar in philosophy to renv or Python virtual environments.
(Of course, you would never encounter this situation if starting from a clean state and using pak with .libPaths() as the default always).

On a less related note: how does it relate to the remotes package? When stumbled upon the original problem, I switched to remotes::install_dependencies() in my workflow, following the current devtools implementation. Will it be deprecated too at some point?

Maybe this https://remotes.r-lib.org/MAINTENANCE.html?q=pak#future-directions answers it?