read_csv produce empty columns when first rows are NA

[artemklevtsov]

Hi.

Thanks for the package. It's really nice.

I have data when first 100+ rows are empty. readr can't parse this cols as expected.

To reproduce:

library(readr)

df <- data.frame(A = rep(c(NA, 1:3), each = 100, times = 2),
                 B = rep(c(NA, "A", "B", each = 200)))

temp_file <- tempfile(fileext = ".csv")
write_csv(df, temp_file)

df2 <- read_csv(temp_file)
head(problems(df2))
#> row col       expected actual
#> 1 101   1 T/F/TRUE/FALSE      1
#> 2 102   1 T/F/TRUE/FALSE      1
#> 3 103   1 T/F/TRUE/FALSE      1
#> 4 104   1 T/F/TRUE/FALSE      1
#> 5 105   1 T/F/TRUE/FALSE      1
#> 6 106   1 T/F/TRUE/FALSE      1

lapply(df, table)
#> $A
#> 
#>   1   2   3 
#> 200 200 200 
#> 
#> $B
#> 
#> 200   A   B 
#> 200 200 200 
lapply(df2, table)
#> $A
#> < table of extent 0 >
#> 
#> $B
#> 
#> 200   A   B 
#> 200 200 200 

[hadley]

Yes, in this scenario you'll need to either specify skip = 100, or manually supply the column types.

[artemklevtsov]

I have file with 50+ cols and about 1k rows. I need not skip a rows because other columns not contains NAs. It's too hard to suuply a types for 50 columns.

As result read_csv can't correctly read the file.