[Rd] warnings() in R-devel: inherits from "warnings" - summary(warnings())

[Martin Maechler]

A few minutes ago, I have committed (svn rev 85445)
to R-devel {R's development source code, https://svn.r-project.org/R/trunk/} 

a change with NEWS entry

    * warnings() now always inherits from "warnings" as documented,
      newly also in the case of no warnings, where it previously
      returned NULL.


In addition to changing the R code to follow the help page,
another plus is that now

  summary(warnings())

in the case of *no* warnings, correctly does

  R> summary(warnings())
  no warnings

instead of in current (released) R 

  R> summary(warnings())
  Length  Class   Mode
       0   NULL   NULL

In a few CRAN / Bioc packages authors had used checks

   if( is.null(warnings()) )

which will no longer do what was intended, as
the condition will always be FALSEm
so we strongly recommend you replace all

    is.null(warnings())
by  length(warnings()) == 0

(which works the same in current *and* future R versions).

One package maintainer has already been alerted, as he had
    stopifnot(is.null(warnings()))
at the end of his test script and needs to replace it.

Something like

   summary(warnings())
   stopifnot(length(warnings()) == 0)

maybe a good way to end *such* test R scripts (where you treat
warnings as errors in the end):
Show (a summary of) the warnings if there are, and still stop.

Otherwise, we now *can* recommend using
  summary(warnings())
more generally,
and particularly useful sometimes after

 options(nwarnings = 1e5) # to get practically all warnings


Best regards,
Martin

--
Martin Maechler
ETH Zurich  and   R Core team