[Rd] attach/detach asymmetry

Barry Rowlingson b.rowlingson at lancaster.ac.uk
Wed May 16 20:54:38 CEST 2007


Is there a good reason why arguments to 'attach' and 'detach' differ so 
much? Yes, I know this is the documented behaviour, it just seems to 
violate the principle of least surprise.

Examples:

  file = "foo.RData"
  df = data.frame(x=1:10,y=1:10)

  attach(file) # attaches the RData file 'foo.RData'
  attach(df) # attaches the object df

  Not surprising in an OO system that a function operating on different 
things does different things. However:

  detach(df)  - works as expected

  detach(file)
gives:
  Error in detach(file) : invalid name

  That's surprise number 1.

  Oh well, lets type the filename out:

  detach("foo.RData")
gives:
  Error in detach("foo.RData") :
          invalid name

  Surprise number 2.

  Check the search() list, and you see that R has stuck 'file:' on the 
start of your file path. So try that:

  detach("file:foo.RData")

  and that works nicely. No surprise there, except perhaps that nothing 
surprising happened.

  So, maybe you can attach() with a 'file:' prefix?

  attach("file:foo.RData")
gives:
  Error in attach("file:foo.RData") :
         file 'file:foo.RData' not found

  - in this case it took it literally as a file name. Surprise number 3.

This is all documented behaviour (actually, I cant find any 
documentation on the 'file:' prefixing), but am I the only person who 
finds it inelegant? Or even downright ugly that the help page for 
detach() gives this example code for attaching and detaching by name?

attach_and_detach <- function(db, pos=2)
      {
         name <- deparse(substitute(db))
         attach(db, pos=pos, name=name)
         print(search()[pos])
         eval(substitute(detach(n), list(n=name)))
      }

  - and that doesn't work for attached data files because of the 'file:' 
prefix.

  I think this could be solved by changing detach(foo) to do something 
different if foo is a character string, and adding the 'file:' prefix 
internally. Or maybe they need rewriting as proper methods?

  Perhaps I'm just being over-sensitive about a matter of aesthetics...

Barry



More information about the R-devel mailing list