na for write_csv would be cool

[ghost]

write_csv(..., na="")

[npjc]

Why not do:

df[is.na(df)] <- ""
write_csv()

IMHO this is clearer to understand at a glance and you keep data output and data mutation as distinct as possible.

[asnr]

That's neat! I didn't know is.na worked sensibly with data frames. However, this approach doesn't seem to play well with factors:

d <- data.frame(x = as.factor(c('foo', NA)))
d[is.na(d)] <- " "
#> Warning message:
#> In `[<-.factor`(`*tmp*`, thisvar, value = " ") :
#>   invalid factor level, NA generated

and it also has the side effect of type casting integer, double and logical columns to character:

e <- data.frame(i = as.integer(c(1, NA)), d = c(0.5, NA), l = c(T, NA))
e[is.na(e)] <- " "
str(e)
#> 'data.frame':    2 obs. of  3 variables:
#>  $ i: chr  "1" " "
#>  $ d: chr  "0.5" " "
#>  $ l: chr  "TRUE" " "

I'm with @ckluss on this one, I think an na option would be helpful--I often get clients requesting specific missing encodings ("", " ", "NULL", but not "NA" so far). @hadley, what do you think? I'd be happy to implement it if you think it's a good idea.

[realAkhmed]

+1.
This feature is very important. Say, PostgreSQL is expecting "\N" as NULLs when importing. MySQL is looking for "NULL" as NULL and so on. One can use this (very smart!) trick suggested by @npjc above but this will coerce types.