Failure to use the already installed Matrix package

[DavisVaughan]

On this GHA build we get the unfortunate error with no extra context: https://github.com/tidymodels/hardhat/runs/3768720403

2021-10-01T15:08:16.0434995Z ##[error]Error: <callr_remote_error: Cannot install packages:
2021-10-01T15:08:16.0446177Z >
2021-10-01T15:08:16.0446592Z  in process 9759 
2021-10-01T15:08:16.0447427Z -->
2021-10-01T15:08:16.0447872Z <simpleError: Cannot install packages:
2021-10-01T15:08:16.0449766Z >
2021-10-01T15:08:16.0475755Z 
2021-10-01T15:08:16.0476201Z  Stack trace:
2021-10-01T15:08:16.0476417Z 
2021-10-01T15:08:16.0476762Z  12. (function (...)  ...
2021-10-01T15:08:16.0477542Z  13. base:::withCallingHandlers(cli_message = function(msg) { ...
2021-10-01T15:08:16.0479420Z  14. get("pkg_deps_internal", asNamespace("pak"))(...)
2021-10-01T15:08:16.0480093Z  15. pak:::pkg_deps_internal2(pkg, upgrade, dependencies)
2021-10-01T15:08:16.0480669Z  16. deps$stop_for_solution_error()
2021-10-01T15:08:16.0481174Z  17. private$plan$stop_for_solve_error()
2021-10-01T15:08:16.0481803Z  18. pkgdepends:::pkgplan_stop_for_solve_error(self, private)
2021-10-01T15:08:16.0483455Z  19. base:::stop("Cannot install packages:\n", msg, call. = FALSE)
2021-10-01T15:08:16.0484128Z  20. base:::.handleSimpleError(function (e)  ...
2021-10-01T15:08:16.0484665Z  21. h(simpleError(msg, call))
2021-10-01T15:08:16.0485370Z  22. base:::stop(e)
2021-10-01T15:08:16.0486781Z  23. (function (e)  ...
2021-10-01T15:08:16.0487017Z 
2021-10-01T15:08:16.0487395Z  x Cannot install packages:

Run locally, we get the following:

> pak::pkg_deps("local::.", dependencies = TRUE)
✔ Updated metadata database: 2.44 MB in 5 files.
✔ Updating metadata database ... done
Error: Cannot install packages:
* local::.: Can't install dependency Matrix
* Matrix: Needs R 3.5.0
* Matrix: Needs R 4.2
Type .Last.error.trace to see where the error occurred

Matrix is a recommended package, which should already be installed, but we currently don't have a way to tell pak to use it and that seems to be related to this issue

[hfrick]

I have a failing built which tries to install the foreign package which is also a recommended package (and now requires 4.0.0): https://github.com/tidymodels/censored/pull/82/checks?check_run_id=3806116333

[gaborcsardi]

@hfrick Why does the same thing work on Linux? https://github.com/tidymodels/censored/pull/82/checks?check_run_id=3806116558#step:5:234

[hfrick]

I was wondering the same thing :sweat_smile:

[juliasilge]

Seeing this on themis as well cc @EmilHvitfeldt

[gaborcsardi]

AFAICT themis needs progressr, which needs R 3.5.0, so even if this issue is solved, you won't be able to install it on R 3.4.x:

❯ pak::pkg_deps_explain("tidymodels/themis", "progressr")
themis -> recipes -> ipred -> prodlim -> lava -> progressr

[gaborcsardi]

@DavisVaughan the same is true for hardhat, seemingly:

> pak::pkg_deps_explain("tidymodels/hardhat", "progressr", dependencies = TRUE)
hardhat -> recipes -> ipred -> prodlim -> lava -> progressr

progressr is a hard dependency of recipes (both CRAN and GH versions), so anything that needs recipes needs at least R 3.5.0:

> pak::pkg_deps_explain("recipes", "progressr")
recipes -> ipred -> prodlim -> lava -> progressr
> pak::pkg_deps_explain("tidymodels/recipes", "progressr")
recipes -> ipred -> prodlim -> lava -> progressr

[gaborcsardi]

@hfrick as for censored, riskRegression is a hard dependency and it needs R 3.5.0:

❯ pak::pkg_deps_explain("tidymodels/censored", "riskRegression")
censored -> pec -> riskRegression

and glmnet is a dev-dependency, and that need R 3.6.0.

[juliasilge]

I'm going to add a reminder here (most of the folks on this thread probably remember) that about a month ago, we had to move recipes to R>=4.0.0 because of fastICA. Anything that tries to build recipes will now need to stick to R>=4.0.0.

EDIT: Wait, that can't be right??? Because themis is now building back to oldrel-3?