[Rd] S3 methods in packages. Change in R 4.0.0?

[Duncan Murdoch]

On 15/11/2019 8:42 a.m., Herbert Braselmann wrote:
> Hello,
> 
> since Bioconductor devel branch 3.11 started, which is operating with
> R-devel 4.0.0 (for e.g. 2019-11-03 r77362 on OS X 10.11.6 El Capitan),
> my package CFAssay 1.21.0 is not built on none of all three Systems,
> Linux, Windos and Mac. The error message refers to xy.coords which is
> called by R function plot.default. I found out that it has to do with
> the S3 object system. In my package I defined an S3 method for R generic
> plot and three for print. To be free from the biological stuff I
> generated here a simple example package:
> 
> https://github.com/ZytoHMGU/myfirstS3steps
> 
> A simple method  'print.my1stClass' is defined there, which writes the
> sumproduct of two numerical entries or numbers of a list, given a class
> attribut matching the name of the method. Up to R 3.6.1 it works as one
> is used to expect from S3 methods. After attaching the package in R
> 4.0.0, a call of 'print' for a list object of that class falls back to
> 'print.default' and writes the list entries. However, when the method is
> sourced, so that it appears in the global environment, it works as
> expected for S3 with a 'print'. Independently thereof, the method is
> shown in the list with .S3methods(print) or .S3methods(call="my1stClass").

You aren't declaring your method as a method in the NAMESPACE file.  I 
hadn't noticed a news item saying that the necessity for this had 
changed, but it is definitely something you should do.  Put this into 
NAMESPACE:

S3method(print, my1stclass)

and it should be fine.

Duncan Murdoch


> 
> A similar problem has function plot, I don't show it here. It leads to
> an error message, when plot.default and xy.coords cannot work with the
> object for which the method was written. I guess that methods for
> 'summary' and all other R base generics, written in a package as I did
> it and attached with the library command would have the same problem in
> R-devel 4.0. I did not try it. My questions are now:
> 
> (1) Is action required or is it a bug in 4.0?
> (2) Should I be more stringent, for e.g. using special directives in the
> NAMESPACE file of the package? I did not.
> (3) Should I change to S4? For e.g. Bioconductor package CGHbase (Mark
> van de Wiel) contains S4 plot-methods for its objects.
> 
> For whom it may be interesting the URL of my Bioconductor package:
> http://bioconductor.org/checkResults/devel/bioc-LATEST/CFAssay/
> 
> Herbert Braselmann
> <http://bioconductor.org/checkResults/devel/bioc-LATEST/aCGH/>
>