[Rd] rdyncall fears removal from CRAN

[Hadley Wickham]

> This is also very relevant for devtools. There have been some recent
> changes to the in-development version of devtools, so that it loads
> development packages into a namespace. Previously, it didn't use
> namespaces, but now that it does, it much more closely approximates
> normal package loading behavior.
>
> It now uses the internal functions registerNamespace and
> unregisterNamespace. Is there a workaround to do these things without
> using the internal functions?

We're very aware that using these internal functions ties us to a
specific version of R, and that they may change at any point in the
future and break our code. That's an acceptable trade-off - using
these internal functions gives us capabilities that makes fixing the
code in the future worthwhile.

A minimal change to R that would satisfy our needs (and I suspect
Daniel's) would be to move makeNamespace into the base package (i.e.
up out of the function it is currently contained in), and to create R
wrapper functions for registerNamespace and unregisterNamespace.  They
wouldn't even need to be exported - we know what we're doing is
dangerous.

If that solution isn't acceptable, we'd also be very happy to work
with any R-core member to help develop a public facing API for
namespace manipulation, including providing tests for common use
cases.

Please let us know what we can do to help!

Hadley

-- 
Assistant Professor
Department of Statistics / Rice University
http://had.co.nz/