How can I make fread read "" as an empty string (i.e. without manually specifying colClasses)? What justifies this fread/fwrite incompatibility and how to avoid it (i.e. is there a way to fwrite so that fread can read empty strings)?

library(data.table)
dt <- data.table(a = 1, b = '')
fwrite(dt, file = 'out.csv')
dt2 <- fread('out.csv')

dt
#    a b
# 1: 1  
dt2
#    a  b
# 1: 1 NA

There are couple of closely related posts (e.g. this one, but it also asks for a trouble by specifying a numeric string). I think my case is much simpler yet more surprising/insidious given somewhat justified expectation that fwrite and fread should be able to handle the same data consistently.

CodePudding user response：

This is a combination of both a previous question and a current bug in data.table#3439 where the empty strings are not recognized correctly.

Two ways to resolve this issue, and perhaps they will not be necessary when #3439 is fixed:

1. ensure that other rows in the b column are non-empty strings; you may not have control over this, so ...

   writeLines(c("a,b", '1,""', '2,"b"'), "out.csv")
   fread("out.csv")
   #        a      b
   #    <int> <char>
   # 1:     1       
   # 2:     2      b
   

2. ensure that the b column is class character:

   ### with the original 1-row CSV file
   fread("out.csv", colClasses=c(b="character"))
   #        a      b
   #    <int> <char>
   # 1:     1       
   

3. When #3439 is resolved, you should be able to just do

   fread("out.csv", na.string=NULL)