fread doesn't check colClasses to be valid type

[mattdowle]

> sapply(fread("A,B\n1,2",colClasses="numeric"),class)   # ok
        A         B 
"numeric" "numeric" 
> sapply(fread("A,B\n1,2",colClasses="integer"),class)    # ok
        A         B 
"integer" "integer" 
> sapply(fread("A,B\n1,2",colClasses="string"),class)      # type "string" works?
          A           B 
"character" "character" 
> sapply(fread("A,B\n1,2",colClasses="c"),class)             
          A           B 
"character" "character" 
> sapply(fread("A,B\n1,2",colClasses="foo"),class)      # should be error that foo isn't valid type?
          A           B 
"character" "character" 
> 

[MichaelChirico]

maybe just produce a warning for all of the above?

[jangorecki]

how exactly we want to define valid type?

[mattdowle]

@jangorecki Looking at Hugh's PR #2545 made me realize the answer to your question. Just as read.csv, "foo" is a valid type to use in colClasses= if as.foo.character() or as.foo.default() exists. Otherwise, standard R error that as.foo method does not exist.

[MichaelChirico]

Whether as.* exists will be dependent on the local user's setup. I suppose this is a desired behavior -- users expecting coercion to foo class with methods defined in bar library can be trusted to handle loading bar themselves before running fread. From a glance at #2545, the backup (to use default interpretation, typically character) with a warning is probably ideal behavior -- if I forget to load bar, I'd rather be warned that the ex-post conversion was abandoned than to have fread fail.