Think about methods for generics in suggested packages

[hadley]

e.g. https://dbplyr.tidyverse.org/reference/pivot_longer.tbl_lazy.html — the tidyr methods look different to the others because (as far as roxygen2 knows), they're not actually methods.

[hadley]

Need to standardise and/or document hack to load packages before roxygenise() is run: https://github.com/tidyverse/dtplyr/blob/master/DESCRIPTION#L42

[gaborcsardi]

There is

Roxygen: list(packages = "foobar")

now. Isn't that good?

[hadley]

I know I tried that for dtplyr and I'm pretty sure it didn't work.

[gaborcsardi]

Is that because you need to attach the package? I am fairly sure that the loading works.

[hadley]

Yeah, that's it.

[gaborcsardi]

We can attach packages then maybe?

[hadley]

Yeah, that should be relatively harmless I'd think.

[hadley]

The original goal of the packages field was to load packages that provided other roclets, but it doesn't seem like attaching these packages will cause harm and creating another field seems like overkill.