Need some way to specify that you only want certain columns

[hadley]

col_types = only("a", "k", "z")

col_types = only("a", "k", z = col_factor(c("a","b")))

??

[HenrikBengtsson]

Since I'm just starting to look into readr, it might be that I'm unaware of some existing features of the package, so please forgive me if that's the case. If not, have a look at how I do it in R.filesets::readDataFrame(). Maybe you can enhance col_types to support named character vectors as well. Example:

readDataFrame(pathname, colClasses=c("*"="NULL", "(x|y)"="integer", "char"="character"))

Here names(colClasses) specified regular expression matching column names. The "*"=NULL specifies that the default column class should be NULL, i.e. to drop/skip all columns by default, except those specified.

In your case I can imagine something like:

read_tsv(pathname, col_types=list("*"=col_skip(), x=col_integer(), y=col_integer(), char=col_character()))

An alternative is to let an empty name represent the default behavior.

You could also extend col_types to also support:

read_tsv(pathname, col_types=c("*"="_", x="i", y="i", char="c"))

such that it expands to the above list. With empty name for default, you'd have:

read_tsv(pathname, col_types=c("_", x="i", y="i", char="c"))

[hadley]

I'm not a big fan of overloading column names with additional structure. What happens if there is a column called *?

[HenrikBengtsson]

That's why I proposed the empty-name alternative. Of course, then there could be empty column names as well. Using regular expressions handles it all, but you'd need to escape.