How to prevent command line args from being interpreted by R vs. only by my script?

Question

I'm using the docopt implementation for R. My script has a command line option where the short form is -g. When I run my script, it seems this argument is first interpreted by R and then by my script. Therefore I get a wrist slap about not specifying a value for the GUI. Can I prevent R from trying to work with these command line args?

Example of a script:

#!/usr/bin/Rscript

suppressPackageStartupMessages(library(docopt))

"docopt practice script

Usage: foo.R [-g <goodies>]

Options:
     -g <goodies>, --goodies=<goodies>  Goodies
" -> doc

opts <- docopt(doc)
cat(sprintf("goodies = %s\n", opts$goodies))

Here's what happens when I run it:

Jennifers-MacBook-Pro-3:scripts jenny$ ./foo.R -g donuts
WARNING: --gui or -g without value ignored
goodies = donuts

If you change the short form of the option from -g to -j, the WARNING goes away … but I have a good reason for using the letter g!

Answer 1

As pointed out by @krlmlr, this issue has to do with Rscript (in your hash bang). One workaround would be to use the functionality provided by the excellent littler in place of Rscript. For example, using #!/usr/bin/Rscript in foo.R, I get the issue:

[nathan@nrussell R]$ ./foo.R -g donuts
WARNING: unknown gui 'donuts', using X11

goodies = donuts

Replacing this with #!/usr/local/bin/r in a new script foo2.R, I get a clean output:

[nathan@nrussell R]$ ./foo2.R -g donuts
goodies = donuts

It looks like you're on an OS X machine, so if you do choose to install littler, just be sure to note the authors' warning:

On OS X, you may want to build it via configure --program-prefix="l" to renamed it to lr as that particular OS thinks R and r are the same

Answer 2

The R and Rscript commands know --args. Compare the output of the following:

R -e "TRUE" --args --silent
R -e "TRUE" --silent

This works due to an early exit if --args is detected. However, the --gui warning is triggered in a separate loop before this.

This means that

Rscript -e "commandArgs()" --args --gui

will work but give the spurious warning, and

Rscript -e "commandArgs()" --gui

gives an error right away. It looks like only --gui and -g are affected.

As a quick-and-dirty hack, one could insert something like

if(!strcmp(*avv, "--args")) {
    break;
}

at the beginning of the GUI-check loop. Until this is changed in R, I suspect there's no choice but to avoid the -g switch or live with the (otherwise harmless) warning.