check_installed() doesn't recognize a load_all()'d package as installed

[DavisVaughan]

> rlang::check_installed("hardhat")
> devtools::load_all(".")
ℹ Loading hardhat
> rlang::check_installed("hardhat")
ℹ The package `hardhat` is required.
✖ Would you like to install it?

1: Yes
2: No

This comes up in a test for hardhat. I have a create_modeling_package() function that is modeled after usethis, it creates a new package and calls use_package("hardhat", type = "Imports") to set up the description file of that package. use_package() eventually calls check_installed("hardhat") and this is failing in my test for create_modeling_package() now.

[DavisVaughan]

Worth noting that an interactive call to devtools::check() passes, but an interactive call to devtools::test() fails. This makes sense to me, because check() tests the built package in another process and doesn't use load_all(), and test() uses load_all() and tests in the current process.

[DavisVaughan]

Oh it is this new commit to fix #1378

5fcde82

The env_is_locked() bit is problematic here (hardhat's env isn't locked while running tests? - I guess because load_all() was used)

[DavisVaughan]

Ah ok so I was surprised that load_all() wasn't sealing the namespace, but it seems like this recent commit in dev pkgload does seal the namespace. r-lib/pkgload@29cc79a

So I think we'd need a pkgload release before the next rlang release for this to be compatible with previous behavior.

# pak::pak(c("r-lib/pkgload", "r-lib/rlang"))

> rlang::is_installed("hardhat")
[1] TRUE
> devtools::load_all(".")
ℹ Loading hardhat
> rlang::is_installed("hardhat")
[1] TRUE

[lionel-]

oh yeah I forgot this was not on CRAN already :(

[lionel-]

I've started the release process of pkgload.