Consider the following, which is new behaviour under R 2.5+:
> df1 <- data.frame( x=2, y='cat')
> df2 <- data.frame( x=3, y='dog')
> rbind( df1[-1,], df2)$y == rbind( df1, df2)[-1,]$y
Error in Ops.factor(rbind(df1[-1, ], df2)$y, rbind(df1, :
Level sets of factors are different
To me this seems illogical; it shouldn't matter whether you remove the first
row of the data.frame before or after augmenting the rows (and didn't, in R
prior to v2.5). And there's more!
> levels( rbind( df1[-1,], df2)$y)
[1] "dog"
> levels( rbind( df1[-1,], df2[-1,])$y)
[1] "cat"
but why should rbind suddenly acknowledge the levels of its first argument in
the second case and not the first?
If the data.frames have more than one row, these issue don't arise; they
occur because (as the documentation says) "The 'rbind' data frame
method first drops all zero-column and zero-row arguments. (If that leaves
none, it returns the first argument with columns otherwise a zero-column
zero-row data
frame.)...". So this behaviour is documented-- but isn't it
nevertheless a bug?
>From the release notes, I gather that the new zero-row-stripping behaviour
was introduced to get round a specific problem with 0*0 data.frames. Given the
above, though, wouldn't it be preferable to just include special code to
deal with the 0*0 case, leaving 0*N cases unaffected?
Mark Bravington
