Feature request: ensure parameter names are valid R identifiers

[Former user]

First, thank you for your useful package. I have a suggestion concerning the naming of the parsed arguments.

In the Unix world, it is a general convention to use the prefices '-' and '--' for short and long option names, respectively. If a long option name consists of multiple words, these are separated by another '-', for example "--end-time". Using your package, I would write:

parser <- arg_parser( 'Example' )
?parser <- add_argument( 
    parser, arg = '--end-time', type = 'double', default = 1e20, 
    help = "Time at which simulation ends"
)
args <- parse_args( parser)

Now, the value of the option is stored in the list 'args' under the name 'end-time'. To access this value, you cannot write

args$end-time    # interpreted as args$end - time

but instead have to write the clumsy

val <- args[['end-time']] 

because words containing '-' are no legal R identifiers. However, it would be much more convenient to write

args$end_time

or something similar. Would it be possible to add an automatic substitution for illegal characters? This could also be made optional. Thanks for your help!

[David Shih]

I've been wanting to do that. It turned out it wasn't a simple fix.

I tried again. I made several lines of changes, and hopefully I didn't break anything. I found no bug after light testing, but more extensive testing will be needed (later).

On the plus side, I found an unrelated bug, and fixed it with the latest commit (it's bad practice, but I am sleepy...)

Things should work as requested in the latest commit.

Pull the latest commit, run roxygen2 and build as detailed in the README. (I don't commit auto-generated files to git because they add bulk.)

Kindly set this issue to resolved if my latest commit solves it.

[Felix Kuehnl]

Thanks for your ultra-fast response ;-) I tried the newest version from your repository and after some short testing it seems to work fine. When adding arguments containing hyphens, these are substituted by underscores. Two things I noticed:

1) You only seem to substitute hyphens ( '-' ) for now. Since you know exactly which characters are allowed in a valid R identifier (letters, numbers, underscores, no numbers in the beginning of identifier) you may want to substitute more aggressively, namely any character that violates the rules. Or is this a bad idea?

2) When throwing an error message about parameters added twice, you should probably print out the substituted form of the parameter name so the user knows that a substitution is going on. For example,

parser <- add_argument( 
    parser, arg = '--n_test1', 
    help = "Testing it till it breaks..."    
)
parser <- add_argument( 
    parser, arg = '--n-test1', 
    help = "Testing it till it breaks..."    
)

will lead to an error that says --n-test1 has already been defined. Without knowing of this subsitution, the user might be confused.

PS: I don't think I can set this issue to resolved since I did not log in when I posted it... sorry!

[David Shih]

Regarding (1), I am not a fan of automatic substitutions, especially R's handling of names with number prefix. I only added the substitution of - to _ because of how common "-" appear in argument numbers. I don't see much utility for other substitutions, and it creates more backwards incompatibility. People who want to use eccentric argument names will just have to live with a more verbose argument indexing syntax.

Even with the - to _ substitution, people who were using the args[['arg-name']] syntax will have their code broken! I documented this incompatibility issue in the README. It is not desirable to emit both arg-name and arg_name, since it adds bloat and complexity.

Regarding (2), I am now providing a more explicit error message.

[David Shih]

Added in 1416348 and 496c615