as.data.frame.function does not return a data frame

[josilber]

The as.data.frame.function (ADFF) function returns a function that can be used to generate a data frame. However, due to ADFF's name it is invoked whenever as.data.frame is called on a function f. This violates the return type specified in ?as.data.frame:

as.data.frame returns a data frame, normally with all row names "" if optional = TRUE.

The reason it's problematic (imo) to have as.data.frame not always return a data frame is that it technically requires library implementers to use is.data.frame every time they invoke as.data.frame to make sure they're now working with a data frame. As an example, consider running the following in a clean R session with no packages loaded (imagine somebody was using df for their data frame but the data failed to load in a previous step, leaving df as the density function for the F distribution):

merge(df, iris)
# Error in as.data.frame.default(x) : 
#   cannot coerce class ""function"" to a data.frame

This looks good -- merge has a helpful error message saying we can't convert df (a function) to a data frame for merging.

Now, consider what happens if we have the plyr package loaded:

library(plyr)
merge(df, iris)
# Error: C stack usage  7969596 is too close to the limit

The merge has failed by infinite looping. The culprit is the merge.default function:

function (x, y, ...) 
merge(as.data.frame(x), as.data.frame(y), ...)
<bytecode: 0x7fa71cb288b0>
<environment: namespace:base>

merge.default is calling as.data.frame on both arguments and then invoking merge; since as.data.frame(x) doesn't return a data frame (as the merge.default implementers assumed it would), merge.default is again being invoked, causing the infinite loop.

Would it be possible to rename as.data.frame.function to something else (e.g. as.df.function) to avoid this?

[andypea]

I'd like to second the request to rename as.data.frame.function. Doing so would have saved me an hour of debugging today.

As @josilber said, it gets called by the as.data.frame generic method, but it doesn't return a data.frame. This silently breaks some of the base R functions.