par(mfg=) and postscript and pdf
This is related to the incorrect bug report PR#7820. Marc Schwartz pointed out in https://stat.ethz.ch/pipermail/r-devel/2005-April/033016.html an example of a real problem. If you call par(mfg=) after par(mfrow) (or mfcol) and before you have done any plotting, NewPage is not called on the device at the start of the first page. That causes the DSC comments to be incorrect on postscript() and much worse on pdf(). You would get the same effect by calling par(new=T) before plot.new() is called, except that is disallowed. par(mfg=) does set new=T internally, and also sets the plot number to a valid one, both of which inhibit calling NewPage. I've managed to overcome this by ensuring that GNewPage always calls NewPage on an unused device (there is an internal 'state' variable which records the latter). This is safe in the sense that the worst it could do is to produce an unwanted new page. I can't see how this could happen but the design of the internals of base graphics is very complicated and undocumented, so I am only putting the change in R-devel (along with fixes to the long-standing graphics bugs PR#1235 and PR#2630).
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