R devel - Sep 2004 - A couple of issues with colClasses/setAs

Consider this:

$ cat test.dat
1 a
2 b

Now, we want to read the 2nd column as a factor and ignore the first
(since it's just a sequential ID). We can't just put "factor"
among
the colClasses (would have been nice), so let's try this instead
> setAs("character","factor",as.factor)
Arguments in definition changed from (x) to (from)
>
read.table("test.dat",colClasses=c("numeric","factor"))
Error in inherits(x, "factor") : Object "x" not found

which is a bit peculiar: Why does it change the argument when that's
going to create a function that doesn't work?? You do need to spell it
out:
>
setAs("character","factor",function(from)as.factor(from))
And now we get somewhere
>
read.table("test.dat",colClasses=c("numeric","factor"))
  V1 V2
1  1  a
2  2  b

but suppose we want to get rid of col.1:
>
read.table("test.dat",colClasses=c("NULL","factor"))
Error in data[[i]] : subscript out of bounds

which looks like a pretty clear bug. In contrast, this works fine
>
read.table("test.dat",colClasses=c("NULL","character"))
  V2
1  a
2  b

so the issue only arises when you have nontrivial coercions.

Presumably, the issue is that the colClasses in those cases
miscalculate indices by forgetting the columns that were skipped.

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard@biostat.ku.dk)             FAX: (+45) 35327907

>From ?read.table (this is about read.table, despite the subject line, I 
believe?)

colClasses: character.  A vector of classes to be assumed for the columns.

"NULL" is not a class in my book (and certainly not one a column can
have).  So no wonder it does not work, and it is not a bug not to work in
undocumented cases.

We can look into making it work, but once you start skipping columns I 
think you should be using scan().  (I also suspect scan did not accept 
NULL when this was implemented.)

On 8 Sep 2004, Peter Dalgaard wrote:
> Consider this:
> 
> $ cat test.dat
> 1 a
> 2 b
> 
> Now, we want to read the 2nd column as a factor and ignore the first
> (since it's just a sequential ID). 
Well, you have to have row names, so that's not actually an advantage.
> We can't just put "factor" among
> the colClasses (would have been nice), so let's try this instead
> 
> > setAs("character","factor",as.factor)
> Arguments in definition changed from (x) to (from)
> >
read.table("test.dat",colClasses=c("numeric","factor"))
> Error in inherits(x, "factor") : Object "x" not found
> 
> which is a bit peculiar: Why does it change the argument when that's
> going to create a function that doesn't work?? You do need to spell it
> out:
> 
> >
setAs("character","factor",function(from)as.factor(from))
> And now we get somewhere
> 
> >
read.table("test.dat",colClasses=c("numeric","factor"))
>   V1 V2
> 1  1  a
> 2  2  b
Might be a good idea to teach colClasses about "factor".
> 
> but suppose we want to get rid of col.1:
> 
> >
read.table("test.dat",colClasses=c("NULL","factor"))
> Error in data[[i]] : subscript out of bounds
> 
> which looks like a pretty clear bug. In contrast, this works fine
> 
> >
read.table("test.dat",colClasses=c("NULL","character"))
>   V2
> 1  a
> 2  b
> 
> so the issue only arises when you have nontrivial coercions.
> 
> Presumably, the issue is that the colClasses in those cases
> miscalculate indices by forgetting the columns that were skipped.
> 
> 
-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595