Skip to content

WishList: Remove Generic Arguments

4 messages · Abby Spurdle, Michael Lawrence, Duncan Murdoch

Original
Abby Spurdle
# 
I apologize if this issue has been raised before.

I really like object oriented S3 programming.
However, there's one feature of object oriented S3 programming that I don't
like.
Generic functions can have arguments other than dots.

Lets say you have an R package with something like:

print.myfunction (f, ...)
{   dosomething (f, ...)
}

Noting that I use function objects a lot.

R CMD check will generate a warning because you've named your object f
rather than x.

I don't want to name my object x.
I want to name my object f.
Naming the object x makes the program unreadable.
Especially if f contains an attribute or an argument named x.

There's a work around.
You can redefine the print function, using something like:

print = function (...) base::print (...)

However, you have to export and document the function.

I think that it would be better if generic functions didn't have any
arguments except for dots.
Michael Lawrence
# 
A generic function is not simply a way to name two functions (methods)
the same. It has a particular purpose, and the argument names are
aligned with and convey that purpose. The methods only implement
polymorphism; they don't change the purpose. Changing the purpose
would make code unreadable.

Michael
On Thu, Aug 9, 2018 at 2:45 PM, Abs Spurdle <spurdle.a at gmail.com> wrote:
Duncan Murdoch
#
On 09/08/2018 5:45 PM, Abs Spurdle wrote:
There are reasons for this requirement.  Suppose I have an object of 
class myfunction named myobj.  However, being a 3rd party, I have no 
idea what class myfunction is about.  Then I might say

print(x = myobj)

That would not work with your method, because the x would be absorbed 
into ..., it would not bind to f.
That's nonsense.  You don't need to name your object the same as an 
argument.  You name your object something readable (e.g. myobj), then 
pass it to print(), which binds it to x.

If x is not a sensible name within the print.myfunction() method, then 
there's a one line fix:

print.myfunction <- function(x, ...) {
   f <- x
   dosomething(f)
}
That's a really, really bad idea.  If there are two generics named the 
same, how are your users going to know which one they are getting when 
they just say print(myobj)?
That makes argument checking much harder.  If a generic always needs two 
arguments, why not name them, so R can complain when a user calls it 
with just one?

Duncan Murdoch
1 day later
Abby Spurdle
# 
there's a one line fix:
Naming the argument x is not an option.
same, how are your users going to know which one they are getting when they
just say print(myobj)?

If redefining the generics is "a really, really bad idea" then an
alternative would be to define a new generic and two methods, one of which
acts as a kind of bridging function, using something like:

myprint = function (...) UseMethod ("myprint")
myprint.myfunction = function (f, ...) {dosomething (f, ...)}

print.myfunction = function (x, ...) myprint (x, ...)
complain when a user calls it with just one?

We can apply that principle to the methods rather than the generic.



On Fri, Aug 10, 2018 at 10:20 AM, Duncan Murdoch <murdoch.duncan at gmail.com>
wrote: