Consider making package load optional

[eddelbuettel]

I usually call roxygenize(roclets="rd") (via this script, usually) which worked fine for many years. It uses roxygenize to translate markup from the R/ directory into content in the man/ directory. This never touches NAMESPACE or DESCRIPTION.

Now, all of a sudden, the function decides to invoke pkgload, forcing a load/build I did not ask for, and seemingly running it with different flags (as I get compiler warnings I do not get otherwise).

I would like to suppress this package loading test. Would you consider a PR to that effect? Maybe skipping if roclets is equal to "rd" so that we do not need a new flag?

(And I checked that this did not happen with version 6.0.1. The code in R/roxygenize.R appears to not have changed so this may be a side effect of other changes.)

[hadley]

This is an unfortunate consequence of the interaction of several packages, but should be much ameliorated with the latest version of pkgbuild, thanks to r-lib/pkgbuild@60162f8#diff-8312ad0561ef661716b48d09478362f3L1

[eddelbuettel]

The diff you reference appears to talk about Makefiles.

I would simply like to get back to the world where roxygenize() never ever attempts to compile (or load) a package. I call it for text substitution.

[eddelbuettel]

I second the comment by @mb706, I am running into the same issue.

The bug is obviously not fixed, so I don't understand why this was closed 17 days ago.

[krlmlr]

In most cases roxygenize() needs to load the code, e.g. to create \usage{} . This has been done previously in an ad-hoc way, now pkgload takes care of this.

Seems that we need better examples for the load_code argument to roxygenize() . Maybe something like this:

env_legacy <- function(path) {
  env <- new.env(parent = ...)
  files <- dir(file.path(path, "R"), full.names = TRUE)
  for (file in files) source(file, local = env)
  env
}

roxygenise(..., load_code = env_legacy)

[eddelbuettel]

Here is another manifestation of the same bug as a fresh SO question:

https://stackoverflow.com/questions/53642735/namespace-option-created-by-rcpparmadillo-package-skeleton-causes-error/

[FlorianSchwendinger]

I see something similar (R-3.5.2 + Debian + roxygen2_6.1.1)

> roxygen2::roxygenise("scs", roclets = "rd")
Loading scs
Error in getDLLRegisteredRoutines.DLLInfo(dll, addNames = FALSE) : 
  must specify DLL via a “DLLInfo” object. See getLoadedDLLs()
Calls: <Anonymous> ... assignNativeRoutines -> getDLLRegisteredRoutines.DLLInfo

interestingly it seems to be connected to the useDynLib setting in the NAMESPACE file.
Using R -e 'roxygen2::roxygenise("scs", roclets = "rd")' plus

useDynLib("scs") 
export("scs", "scs_control")

does work. However changing the NAMESPACE file to

useDynLib("scs", .registration = TRUE)
export("scs", "scs_control")

then I get the error from above.

Hope this helps to narrow down the source of the issue.

[TWilliamBell]

I believe I've encountered the same problem due to the useDynLib line not being generated in the namespace on a package using Rcpp.

[eddelbuettel]

Yes -- it is very much still an open issue and regression. It bit students in my course trying to a submit a package containing a simple Rcpp function -- some were hit by this just last evening.

[Inferrator]

Issue still present in 6.1.1