windows testing

1. Tried running "test-all-extras" today.  It failed comparison on the
last of the regression tests, "reg-plot".  As far as I could tell, the
output looked okay.  The differences were in reported values of the
relative range of values, as in

for(j.fac in 1e-12* c(10, 1, .7, .3, .2, .1, .05, .03, .01)) {

+ ##           ====
+     #set.seed(101) # or don't
+     x <- pi + jitter(numeric(101), f = j.fac)
+     rrtxt <- paste("rel.range =", formatC(relrange(x), dig = 4),"* EPS")
+     cat("j.f = ", format(j.fac)," ;  ", rrtxt,"\n",sep="")
+     plot(x, type = "l", main = rrtxt)
+     cat("par(\"usr\")[3:4]:", formatC(par("usr")[3:4], wid = 10),"\n",
+         "par(\"yaxp\") :   ", formatC(par("yaxp"), wid = 10),"\n\n", sep="")
+ }
j.f = 1e-11 ;  rel.range = 553.9 * EPS
par("usr")[3:4]:     3.142     3.142
par("yaxp") :        3.142     3.142         3

j.f = 1e-12 ;  rel.range = 56.02 * EPS
Warning: relative range of values = 85.307 * EPS, is small (axis 2).
par("usr")[3:4]:     3.142     3.142
par("yaxp") :        3.142     3.142

and some warnings about character widths:

Warning: font width unknown for character 144
Warning: font metrics unknown for character 144
Warning: font width unknown for character 144
Warning: font width unknown for character 157
Warning: font metrics unknown for character 157
Warning: font width unknown for character 157

2.  In test-Internet, the file comparisons were different (in what
looked like trivial ways: the directory and list from CRAN was
different).  The test produced a file "internet.Rout2" that had no
corresponding .save file.

Duncan