What '.' Means in R, and Why it Matters

As far as I can tell, the R community has no generally-accepted style guide. Google and Hadley Wickham both have style guides, but across and even within CRAN packages, different naming and spacing conventions abound. You’re likely to find variables named in camelCase, snake_case, or, interestingly, dot.case. This last convention is unusual, because unlike many languages, R does not enforce specific syntactic meaning for dots. Dots can denote methods for S3 classes, but they don’t have to. This means that R only cares about dots sometimes, with confusing results.

S3 generic functions, like print, use the function UseMethod to dispatch the appropriate method for the data type. The methods are named using the construction function.class (i.e., print.lm, print.data.frame). Sometimes you can even call the methods directly, without the generic. For example:

> print('hi')
[1] "hi"
> print.data.frame('hi')
<0 rows> (or 0-length row.names)

The character object "hi" isn’t a data.frame, but R will try to call it like a normal function anyway. S3 is very unstructured in this way. The problem is that R can’t always distinguish between S3 generics/methods and dot case functions:

> example <- function(x) print(mean(x))
> example.two <- function(x) print(sum(x))
> methods(example)
[1] example.two
see '?methods' for accessing help and source code
Warning message:
In .S3methods(generic.function, class, parent.frame()) :
function 'example' appears not to be S3 generic; found functions that look like S3 methods 

Here I’ve defined two unrelated functions, but R isn’t sure what to do with them. example isn’t an S3 generic, because I don’t call UseMethod to dispatch example methods for different classes, but because of S3 naming structure, example.two looks like an S3 method on the class two. S3 is so informal that there is no checking to distinguish between the possibilites, and the methods() function lists example.two as a method (although at least it warns you that it’s not sure).

This behavior is both unexpected and problematic. If you’re relying on a specific behavior from methods(), someone else’s package (or your own functions!) could give you unreliable results. There is no reason to use dot case except as convention, and it’s not a set convention, even in base R. There are even base R functions that use a different case for the function name and arguments (colSums (x, na.rm = FALSE, dims = 1))! I’d love to see an R overhaul, where a new R version is released with a consistent style and an accompanying style guide. But until that time, R users should at least stop using dot case.

Done At: Dec 10,2015

Posted with : R