Import only roclet

[hadley]

When you add a new method to a generic defined in another package, you need to first import that generic. This makes it easy to get stuck in a state (e.g. https://github.com/ThierryO/testS4) that you can't easily get out without modifying the NAMESPACE by hand.

One solution would be to have a separate roclet that works like collate. I think the translation from all the import directives is straightforward, so this should work. (But it can't do all namespace directives because @export definitely needs the loaded code)

[richierocks]

I'm currently working around this issue with the script here:

https://gist.github.com/richierocks/21b29f13320716ba6fca

The code isn't yet robust enough to go into roxygen, but works well enough for people that need a quick fix.

cc: @ThierryO

[hadley]

This will require:

• A C++ function, find_imports() analogous to find_includes() that parses all R files and extracts any tag starting with imports or equal to useDynLib()

• An update_imports() function analogous to update_collate(), which writes out a simple NAMESPACE listing all imported functions (provided any exist, and adding the correct comment so roxygen can later overwrite)

• Updates to roxygenise() and devtools::document() to collate, update imports, then roxygen. Given that pkgload now exists, it may now be possible to move more of the logic in to roxygenise so that document() becomes a thinner wrapper

[hadley]

I'm now thinking that it will be better to re-use the parse data that we have to generate later on. I think it will be possible to split the process into two pieces - one that works purely with parse data, and the other that also uses the package environment.