[Rd] trace() an S3-classed function in .GlobalEnv

[Michael Chirico]

Thanks, that seems like a reasonable assessment.

Also worth noting that tracing wipes out a function's attributes,
which is also not documented:

foo = function(x) {
  invisible(1 + 1)
}
attr(foo, "bar") <- 2

trace(foo, quote(message('traced')))
names(attributes(foo))
# [1] "original" "source"   "class"

PS one day I hope to master the dark art of choosing r-devel or
bugzilla for issues :)

On Wed, Jun 8, 2022 at 3:44 AM Martin Maechler
<maechler using stat.math.ethz.ch> wrote:
>
> >>>>> Michael Chirico
> >>>>>     on Mon, 6 Jun 2022 23:09:29 -0700 writes:
>
>     > Consider:
>
>     > foo <- function() {
>     > invisible(1 + 1)
>     > }
>     > trace(foo, quote(message("traced")), print=FALSE)
>     > foo()
>     > # traced
>
>     > But with a minimal change:
>
>     > class(foo) <- "bar"
>     > trace(foo, quote(message("traced")), print=FALSE)
>     > # Constructing traceable class “barWithTrace”
>     > # Error in .classEnv(className) :
>     > #   unable to find an environment containing class “bar”
>
>     > I don't see anything like this mentioned in ?trace (nor does a Google
>     > search turn up more than a handful of references to this error),
>     > and from trying to debug what trace() is doing, we arrive to the error line[1]:
>
>     > .makeTraceClass(traceClass, class(original))  #
>     > methods:::.makeTraceClass("barWithTrace", "bar")
>
>     > I don't quite follow what's going on here, but it looks like trace()
>     > is trying to determine an S4 class definition for "bar", but isS4(bar)
>     > is FALSE.
>
>     > I can (apparently -- not sure if there are as yet-unseen downstream
>     > consequences) work around the issue by unsetting the class, tracing,
>     > then re-setting:
>
>     > class(foo) <- NULL
>     > trace(foo, quote(message("traced")), print=FALSE)
>     > class(foo) <- "bar"
>
>     > But obviously this is a bit clunky. Is this a bug, or am I missing something?
>
> Just a short note of  semi-confirmation:
>
> At the time S4 generics and methods were introduced into R,
> trace() was made much more flexible, notably to be able to trace
> S4 methods.
>
> It can well be that it originally also worked for functions with
> an explicit S3 class, but as such functions are very rare, it
> could well be you've found a bug, namely that trace() assumes
> that if a function has a non-trivial class, it must be an S4
> one.
>
> ... and I know you know how to report bugs ;-)
>
> Thank you in advance!
> Martin
>
>     > Mike C
>
>     > [1] https://github.com/r-devel/r-svn/blob/e2a64a4e14adbc4e9e8635eaa8cbd2835ce1d764/src/library/methods/R/trace.R#L240
>
>     > ______________________________________________
>     > R-devel using r-project.org mailing list
>     > https://stat.ethz.ch/mailman/listinfo/r-devel