Confusion over "standalone" use cases for C symbol dependent functions (i.e. rlang "types-check")

[jimbrig]

Overview

Vendoring rlang's standalone "types-check" helpers with usethis::use_standalone() requires import(rlang) in NAMESPACE due to C symbol registration.

Example

When using usethis::use_standalone("r-lib/rlang", file = "types-check") to vendor rlang's argument/type checking helpers (e.g., check_bool()), the resulting functions fail at runtime unless the host package includes import(rlang) in its NAMESPACE.

Using only importFrom(rlang, check_bool) or related partial imports does not register the required C-level backend symbols (e.g., ffi_standalone_is_bool_*), resulting in errors like:

Error in check_bool(FALSE) : object 'ffi_standalone_is_bool_1.0.7' not found

This seems to defeat the primary appeal of the "standalone" vendoring model for packages that want to minimize imports and namespace pollution.

Steps to Reproduce

1. In a package, run:

usethis::use_standalone("r-lib/rlang", file = "types-check")

2. Call check_bool(FALSE) in your package:

pkgload::load_all()
check_bool(FALSE)

Observe the error about the missing C symbol unless you add import(rlang) as a full import.

Note

The error will occur regardless if the underlying function being tested, i.e. check_bool is imported in NAMESPACE or not. Only full import(rlang) works.

Expected

Vendored "standalone" helpers—especially those intended for minimal-dependency usage—should not require a full import(rlang). Or, if they now require C symbols, documentation should clearly reflect this required import, so package authors understand that vendoring is only an API stability convenience, not a dependency reducer for these functions.

Actual

Full import is required for runtime C symbol registration. Partial imports or vendoring alone are insufficient due to the C backend of the helpers. This leads to confusing "object ... not found" errors for consumers following minimal dependency best practices.

Questions & Suggestions

Could usethis (or rlang standalone docs) clarify that full import(rlang) is required for all C-backed standalone helpers?

Is there a possibility for the vendored R wrappers to check import status at runtime and error clearly, or for usethis to warn during vendoring?

Should this limitation/warning be more prominently featured in the templates or documentation?

[jimbrig]

Apologies as I just came across Issue #1853 so this can be ignored, however, a more directed approach to "standalone" usage patterns for package developers would be a great enhancement!

[DavisVaughan]

Doesn't use_standalone() automatically bring in the Import statement for rlang for you?

Either way, yea this one from rlang is standalone-ish, but not truly standalone (performance reasons), but we haven't worried too much about that.