Revisit definition of slice(df)

[krlmlr]

Currently returns all rows, but it's not tested anywhere. I think it should return an empty data frame instead, and slice(df, i1, i2) should be the same as slice(df, c(i1, i2)).

@hadley: Thoughts?

[hadley]

I think the motivation was for filter(x) and slice(x) to give identical results, but maybe that's wrong because an empty logical should be treated as selecting all rows, whereas an empty integer should be treated as selecting no row.

Equally, filter() ANDs together the inputs, and here you're proposing (effectively) that slice OR's together the inputs.

[krlmlr]

Yes, that's the way I'd see it -- perhaps not even OR, but UNION ALL. FWIW, slice(df, i1, i2) currently throws an error.

[hadley]

Can you tackle this one too please? I think it should be a one-line change in slice_impl().

[krlmlr]

I have now found the test. It seems that slice() intentionally accepts either 0 or one args, negative values are treated as exclusions. The current behavior actually makes some sense, I see no reason to change it.

We could still consider supporting multiple arguments to slice(), but only after #2311.