Inconsistent behaviour with read_csv and skip > #rows

[coolbutuseless]

1. When col_types is specified and skip is equal to or greater than the number of actual rows, read_csv() returns a data.frame with 1 row.
2. When col_types is not specified and skip is equal to or greater than the number of actual rows, read_csv() throws an error.

I think Case1 is wrong to return a row of results when there aren't any, and should probably return a zero-row data.frame.

EDIT: Case 2 is handled OK i.e. If column types aren't specified, and there are no rows from which to infer type, you can't really return anything sensible.

I found this inconsistency when doing chunked reads from a large CSV file, and a zero-row data.frame was going to be an indicator that I'd run out of data.

> read_csv("1,2\n3,4", col_names=c('a', 'b'), col_types='ii')
Source: local data frame [2 x 2]

  a b
1 1 2
2 3 4
> 
> read_csv("1,2\n3,4", col_names=c('a', 'b'), col_types='ii', skip=2)
Source: local data frame [1 x 2]

   a  b
1 NA NA
> 
> read_csv("1,2\n3,4", col_names=c('a', 'b'), skip=2)
Error: You have 2 column names, but 0 columns

[coolbutuseless]

This bit me again recently, so I boiled it down to an even simpler issue.

Issue: readr returns a single row of NA data when there are no rows in a CSV file.

To reproduce:

• sh echo "x,y" > buggy.csv
• readr::read_csv("buggy.csv", col_types="ii")
• Expected result: An empty data frame with zero rows.
• Actual result: A data.frame with a single row of NA values.

[hadley]

If you don't supply the types, do you think it's best to return the most restrictive type (i.e. logical) or the least restrictive type (i.e. character)? I think that's better behaviour than throwing an error (esp. since it's useful to do read_csv(..., n_max = 0) to get just the column names)

[hadley]

Here's what I have now

read_csv("a,b\n1,2")
#> Source: local data frame [1 x 2]
#> 
#>       a     b
#>   (int) (int)
#> 1     1     2
read_csv("a,b\n1,2", c("a", "b"), "ii", skip = 2)
#> Source: local data frame [0 x 2]
#> 
#> Variables not shown: a (int), b (int)
read_csv("a,b\n1,2", c("a", "b"), skip = 2)
#> Warning: 1 parsing failure.
#> row col    expected      actual
#>  --  -- 0 col names 2 col names
#> Source: local data frame [0 x 0]
read_csv("a,b\n1,2", skip = 2)
#> Source: local data frame [0 x 0]
read_csv("a,b\n1,2", n_max = 0)
#> Source: local data frame [0 x 2]
#> 
#> Variables not shown: a (int), b (int)
read_csv("a,b\n")
#> Warning: 1 parsing failure.
#> row col    expected      actual
#>  --  -- 0 col names 2 col names
#> Source: local data frame [0 x 0]

That seems reasonably consistent to me

[hadley]

Hmmm, I think the main thing missing is that if you have column names, but no data and no column types, you get a 0 x 0 data frame - that's not quite right. I've tweaked it to make sure there are always enough col types, using character to pad out:

read_csv("a,b\n1,2")
#> Source: local data frame [1 x 2]
#> 
#>       a     b
#>   (int) (int)
#> 1     1     2
read_csv("a,b\n1,2", c("a", "b"), "ii", skip = 2)
#> Source: local data frame [0 x 2]
#> 
#> Variables not shown: a (int), b (int)
read_csv("a,b\n1,2", c("a", "b"), skip = 2)
#> Source: local data frame [0 x 2]
#> 
#> Variables not shown: a (chr), b (chr)
read_csv("a,b\n1,2", skip = 2)
#> Source: local data frame [0 x 0]
read_csv("a,b\n1,2", n_max = 0)
#> Source: local data frame [0 x 2]
#> 
#> Variables not shown: a (int), b (int)
read_csv("a,b\n")
#> Source: local data frame [0 x 2]
#> 
#> Variables not shown: a (chr), b (chr)

[coolbutuseless]

Thanks! This looks great!