[Rd] Class union of custom classes

[pik@pp@@devei m@iii@g oii gm@ii@com]

Hi Ezra,

thanks, this was very helpful! Your answer got me thinking, and I have tried a couple of more approaches. I thought it would be good to document them here in case someone stumbles on this issue in the future.


I have tried to define classes with combatible initializers:

> # test 5
+ setClass("foo", contains = "logical")
+ setClass("bar", contains = "logical")
+ setClassUnion("foobar", c("foo", "bar"))
+ setClass("withId", contains = "foobar", slots = c(id = "character"))
+ w1 <- new("withId", new("foo", TRUE), id = "test 5")
> Error in initialize(value, ...) : 
  'initialize' method returned an object of class “foo” instead of the required class “withId”

> # test 6
+ setClass("foo", slots = list(x = "logical"))
+ setClass("bar", slots = list(x = "logical"))
+ setClassUnion("foobar", c("foo", "bar"))
+ setClass("withId", contains = "foobar", slots = c(id = "character"))
+ w1 <- new("withId", new("foo", x = TRUE), id = "test 6")
> > Error in initialize(value, ...) : 
  'initialize' method returned an object of class “foo” instead of the required class “withId”

I have also tried to "trick" setClassUnion by naming members of foo and bar .Data:

> # test 7
+ setClass("foo", slots = list(.Data = "logical"))
+ setClass("bar", slots = list(.Data = "logical"))
+ setClassUnion("foobar", c("foo", "bar"))
+ setClass("withId", contains = "foobar", slots = c(id = "character"))
+ w1 <- new("withId", new("foo", .Data = TRUE), id = "test 7")
Error in initialize(value, ...) : 
  'initialize' method returned an object of class “foo” instead of the required class “withId”

The approach you proposed is the only one that works, but with a catch. The withId class will not have access to foo methods, and wrappers are additionally needed to imitate inheritance. If foo has only a few methods, this approach is maintainable. For now, this is the approach I will follow in the real problem I am dealing with. 

Kind Regards,
Pantelis

-----Original Message-----
From: Ezra Tucker <ezra using landtucker.com> 
Sent: Sunday, July 3, 2022 7:57 PM
To: pikappa.devel using gmail.com; r-devel using r-project.org
Subject: Re: [Rd] Class union of custom classes

Hi Pantelis,

What usually helps me in these kinds of puzzles is splitting out
(mentally) the s4 part from the s3 part. The first test you mention, using the class "withId" has an s3 part of a "maybeNumber" (numeric or
logical) and an s4 part of a slot "id". Kind of hidden will be a second slot, ".Data" which contains the s3 data --The w1 value you get is essentially a numeric, and will be subject to numeric methods (ie you could do w1 + 2) and it'll add 2 to all of the values in the .Data slot.

Test 2:
part of the reason setClassUnion works in Test 1 is because the member classes of withID have the same slots-- both of them are .Data.
However, the definition of "foo" in this test has one slot, xb, which is logical, and no .Data slot/no s3 part. foo and withId have incompatable initializers (the "initialize" method for their respective classes so that's why you're seeing this error.

Test 3:
In general, I'd probably avoid using setIs, as it sets an explicit relationship between two classes, whether or not it's logical to do so.
While the initializer for "maybeNumber" ought to complete, because of the issues raised above about Test 2, it'll prevent the object from being created.

Test 4:
This one's tricky. A revealing question is, what happens when you try

> w1 <- new("withId", TRUE, id = "test 4")

Error in initialize(value, ...) : 
  cannot use object of class "logical" in new():  class "withId" does not extend that class

wheras if you did

> setClass("withId", slots = c(data = "maybeNumber", id = "character"))
> w1 <- new("withId", data = new("foo", TRUE), id = "test 4")

it should work properly (or you could do data = 1.2 - it won't recognize a value TRUE as being of class foo since it's logical. Your withId has to have .Data of class numeric or foo, which themselves have incompatible initializers.

Hopefully this helps!

-Ezra

On Sun, 2022-07-03 at 13:02 +0200, pikappa.devel using gmail.com wrote:
> Dear all,
> 
>  
> 
> This code, mostly copied from the setClassUnion help page, works as 
> described in the documentation:
> 
>  
> 
> # test 1
> 
> setClassUnion("maybeNumber", c("numeric", "logical"))
> 
> setClass("withId", contains = "maybeNumber", slots = c(id =
> "character"))
> 
> w1 <- new("withId", 1.2, id = "test 1")
> 
>  
> 
> However, the following three tests do not work:
> 
>  
> 
> # test 2
> 
> setClass("foo", slots = list(xb = "logical"))
> 
> setClassUnion("maybeNumber", c("numeric", "foo"))
> 
> setClass("withId", contains = "maybeNumber", slots = c(id =
> "character"))
> 
> w1 <- new("withId", 1.2, id = "test 2")
> 
>  
> 
> # test 3
> 
> setClass("foo", slots = list(xb = "logical"))
> 
> setClassUnion("maybeNumber", c("numeric"))
> 
> setIs("foo", "maybeNumber")
> 
> setClass("withId", contains = "maybeNumber", slots = c(id =
> "character"))
> 
> w1 <- new("withId", 1.2, id = "test 3")
> 
>  
> 
> # test 4
> 
> setClass("foo", contains = "logical")
> 
> setClassUnion("maybeNumber", c("numeric", "foo"))
> 
> setClass("withId", contains = "maybeNumber", slots = c(id =
> "character"))
> 
> w1 <- new("withId", 1.2, id = "test 4")
> 
>  
> 
> All three return:
> 
>  
> 
>   Error in initialize(value, ...) : 
> 
>       'initialize' method returned an object of class "numeric"
> instead of
> the required class "withId"
> 
>  
> 
> The error comes from:
> 
>  
> 
> traceback()
> 
> 3: stop(gettextf("'initialize' method returned an object of class %s 
> instead of the required class %s",
> 
>        paste(dQuote(class(value)), collapse = ", "), 
> dQuote(class(.Object))),
> 
>        domain = NA)
> 
> 2: initialize(value, ...)
> 
> 1: new("withId", 1.2, id = "test 2")
> 
>  
> 
> I would expect tests 2-4 to work similarly to the first test. Is the 
> above error the intended behavior of setClassUnion? I do not see 
> anything that would prevent this in the documentation. Is there 
> something I am missing here?
> 
>  
> 
> Any help would be very much appreciated!
> 
>  
> 
> Kind regards,
> 
> Pantelis
> 
> 
>         [[alternative HTML version deleted]]
> 
> ______________________________________________
> R-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel