[R-pkg-devel] Deprecating apparent S3 method, changing name

[Ivan Krylov]

В Wed, 25 Sep 2024 05:13:31 +0000
Murray Efford via R-package-devel <r-package-devel using r-project.org> пишет:

> When I deprecate the old functions (by exporting a shell function
> that calls .Deprecated and the new function), I get a package check
> note for e.g. esa.plot "* checking S3 generic/method consistency ...
> NOTE Apparent methods for exported generics not registered: etc."
> 
> This is understandable, but how can I avoid it while properly
> deprecating the old functions?

R CMD check has a hard-coded list of functions that should not be
considered S3 methods, in tools:::nonS3methods(). All other functions
that start with the name of an exported generic followed by a dot will
be considered unregistered S3 methods.

Theoretically speaking, if there was a place in your package where you
called the esa() generic with a user-supplied object, esa.plot() would
be a reachable S3 method, because the user would be able to construct
structure(list(), class = 'plot') and give it to your package code to
call esa(plot_object) and have it dispatched to esa.plot(). (Which
wouldn't be useful to the user, but that's how S3 works.)

I'm afraid that the NOTE is unavoidable in a package that has both an
esa() generic and an esa.plot() non-method function. Perhaps it could
be explained in the submission comment? (Especially since esa.plot()
currently seems unreachable as an S3 method to normal users of the
package.)

I wonder if it's possible to set aside a "please don't consider me for
S3 dispatch" bit in the CLOSXP objects. Then package authors would be
able to set it on their unfortunately named functions, R CMD check
would look it up before considering them to be unexported methods, and
the function findFunWithBaseEnvAfterGlobalEnv() that's responsible for
looking up non-registered S3 methods during UseMethod() would skip over
such closures.

-- 
Best regards,
Ivan