do() does not work with inlined function definitions

[holgerbrandl]

Using todays github-snapshot this works

identity <-function(x) x
iris %>% group_by(Species) %>% do(identity(.))

but the the same logic with the method definition being inlined fails without any obvious reason

iris %>% group_by(Species) %>% do(function(x){ x }(.))
> Error: Results are not data frames at positions: 1, 2, 3

It looks like it worked a while ago (see http://stackoverflow.com/a/22182914/590437)

Adding another set of round brackets around the function definition seems to solve the problem, but I doubt that many users will guess this:

iris %>% group_by(Species) %>% do((function(x){ x })(.))

For improved readability it would be great if (.) could be optionally left out. Then function object would be passed as argument to do which would need to invoke it with the the group table as sole argument. This was how it worked beautifully with ddply.

[kevinykuo]

@holgerbrandl you should really enclose anonymous functions in parentheses.

possibly related: tidyverse/magrittr#12

[hadley]

The problem is that function(x){ x }(.) parses as function(x)({ x }(.)), not (function(x){ x })(.). I don't think this has changed recently (that SO question is the older do() interface).

The whole point of the . pronoun is to avoid defining anonymous functions. Maybe what you're asking for is for this to work?

iris %>% group_by(Species) %>% do(identity)

[holgerbrandl]

@hadley: Indeed that's what I would love to do! It's the equivalent of ddply(iris, .(Species), identity). However, it does not work with the current version of dplyr:

iris %>% group_by(Species) %>% do(identity)
> Error: Results are not data frames at positions: 1, 2, 3

@KyKuo: I just call them right away with (.) because I was under the impression that this is the only way to go when using do(). It looks odd to me too.

Personally, I think that anonymous functions should be possible because when using do(). They allow for an optimal reading of code without jumping around. E.g

data %>% mutate(...) %>% group_by(...) %>% do(function(x){ ... really fancy stuff ..}) %>% select ...

compared to

never_reused_fun_with_odd_name <- function(x){ ... really fancy stuff ..}
data %>% mutate(...) %>% group_by(...) %>% do(never_reused_fun_with_odd_name(.)) %>% select ...

[hadley]

@holgerbrandl but you don't need an anonymous function to do that:

data %>% mutate(...) %>% group_by(...) %>% do({
  ... really fancy stuff ...
}) %>% select ...

[holgerbrandl]

@hadley That's pretty cool and indeed makes anonymous function unnecessary here. Thank you very much.