col_types_concise(): collector for "unknown" type, cf. colClasses=NA

[HenrikBengtsson]

I'm trying out readr and I like what I see :)

I'm missing a way to specify "I let you" choose the best class for the column, cf.

> cat("a\t2\t1.2", file="foo.tsv")

> str(read.table("foo.tsv", colClasses=c("character", "integer", NA)))
'data.frame':   1 obs. of  3 variables:
 $ V1: chr "a"
 $ V2: int 2
 $ V3: num 1.2

> str(readr::read_tsv("foo.tsv", col_names=FALSE))
Classes 'tbl_df', 'tbl' and 'data.frame':       1 obs. of  3 variables:
 $ X1: chr "a"
 $ X2: int 2
 $ X3: num 1.2

> str(readr::read_tsv("foo.tsv", col_names=FALSE, col_types="cid"))
Classes 'tbl_df', 'tbl' and 'data.frame':       1 obs. of  3 variables:
 $ X1: chr "a"
 $ X2: int 2
 $ X3: num 1.2

but ideally also something like:

> str(readr::read_tsv("foo.tsv", col_names=FALSE, col_types="ci?"))
 $ X1: chr "a"
 $ X2: int 2
 $ X3: num 1.2

PS. I picked '?' as in R.utils::colClasses("ci?") == c("character", "integer", "NA"). Related and minor: Did you consider using another symbol than underscore for skipping a column, because it becomes hard to count when there are several of them in a row, e.g. c______i v. c------i.