Print method for subclass makes := return visible output

[eliocamp]

I'm extending data.table class and I'm having a problem. When defining a custom print method operations with := print the object instead of returning it invisibly.

library(data.table)

dt <- data.table(x = 1)

dt[, y := 1]  # No output. Correct!

setattr(dt, "class", c("someclass", class(dt)))

dt[, y := 1] # still no output. Yey!

print.someclass <- function(x, ...) {
  NextMethod("print")
  # maybe other things
}

dt[, y := 1]   # there's output, Noooo!

(The output doesn't seem to appear when reprexed, but it's there in interactive mode, please test)

I don't know what I'm doing wrong and I cannot find much documentation on this.

[MichaelChirico]

@mattdowle I don't quite get what all is meant to be covered here:

length(SYS) > 3L && is.symbol(thisSYS <- SYS[[length(SYS)-3L]][[1L]]) &&
          as.character(thisSYS) %chin% mimicsAutoPrint

First thing that comes to mind is to look for NextMethod in SYS in addition to the mimicsAutoPrint... Or is there some way for downstream users to expand mimicsAutoPrint (on their own? Through the Issues tracker?)

[nikosbosse]

We're seeing a similar issue in our R scoringutils. Is there any solution or steps we could take? Thank you!

[ben-schwen]

First thing that comes to mind is to look for NextMethod in SYS in addition to the mimicsAutoPrint... Or is there some way for downstream users to expand mimicsAutoPrint (on their own? Through the Issues tracker?)

I don't think that expanding mimicsAutoPrint with a curated list for every custom printing is suitable in the long run. Looking for NextMethod on second to last position in addition with length(class(dt)) seems like a viable approach to me

[MichaelChirico]

Looking for NextMethod on second to last position in addition with length(class(dt)) seems like a viable approach to me

Still feels pretty clunky to me -- does that distinguish explicit print(dt) on a class extension?

edit: no, it won't, c.f.:

print(dt)

# yields
sys.calls()
# [[1]]
# print(dt)

# [[2]]
# print.someclass(dt)

# [[3]]
# NextMethod("print")

# [[4]]
# print.data.table(dt)

It looks like the key to detecting the autoprint is this esoteric first entry in the call stack:

sys.calls()
# [[1]]
# (function (x, ...) 
# UseMethod("print"))(x)
# ...rest of stack...

That's just the print() generic being invoked in a substitute()-y way:

substitute(f(x), list(f = print))
# (function (x, ...) 
# UseMethod("print"))(x)

identical(sys.calls()[[1]][[1]], print)
# [1] TRUE

That identical() condition looks pretty good to me as a test of "auto-printed in interactive session" -- in particular explicit print() does not pass that, as sys.calls() looks like so in that case:

sys.calls()
# [[1]]
# print(dt)

# [[2]]
# print.data.table(dt)

Another idea is to add an argument to print.data.table() like suppressPrint=FALSE that downstreams could run to force print suppression, sent along through NextMethod().

[MichaelChirico]

Hmm, is it considered intentional behavior that

print(as.data.table(mtcars)[,y:=2])

Does not print anything? I would guess no?

[eliocamp]

print(as.data.table(mtcars)[,y:=2]) is equivalent to as.data.table(mtcars)[,y:=2], since results are printed implicitly in the console, right? Since the expected behaviour is for := to suppress printing, then, yeah, it is intended.

[aitap]

The idea in #6631 is that what R does in order to auto-print as.data.table(mtcars)[,y:=2] is not exactly print(as.data.table(mtcars)[,y:=2]), but more like eval(substitute(print(x), list(print = base::print)), list(x = previously_obtained_value), globalenv()), and having the contents of base::print in the call stack instead of just quote(print) can be detected. The exact code is from R-3.5...3.6, but the behaviour before that was the same and dates back to R-3.2.0, so we should be fine depending on R ≥ 3.3.