Recent changes to R-devel has altered the way arguments are matched in
most primitives to be identical to the argument matching in
interpreted functions. So if argument names are supplied they are
checked, surplus arguments are errors, and it is possible to use
(perverse) constructions like substitute(e=list(x=1), y ~ x).
There are a few deliberate exceptions: functions like
call switch .C .Call
need to always match the first argument avoiding partial matching to
named arguments in '...', and do use positional matching. And the
perverse
x <- 1
x <- `length<-`(value=3, x=x)
is now an error: two-argument replacement functions still need 'value'
supplied as the second argument but the names are checked.
This has caught long-term errors in several packages, e.g. people
writing proc.time(<expression>) where system.time() was more
plausible, misusing nargs() (if you want the number of args in ...,
use length(list(...)), not nargs(...)), using on.exit(par=opar) not
on.exit(par(opar)), and so on.
Specifically CRAN packages NMF, RobAStBase, TTR, aroma.light, catspec,
gvlma, mvbutils, orloca, orloca.es, sm and staRt now fail in R-devel
(and also some that use them).
--
Brian D. Ripley, ripley at stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272866 (PA)
Oxford OX1 3TG, UK Fax: +44 1865 272595
