ggplot2 code that works interactively (rkward) crashes under Lyx/pgfSweave - Hints please!

Question

I encounter the following error in the *.Rnw.log:

Error:  chunk 3 (label=printgraph) 
Error in grid.Call.graphics("L_text", as.graphicsAnnot(x$label), x$x,  : 
  Metric information not available for this family/device 

only in Lyx and depending upon the statement

+ scale_y_log10(ylim=c(0.025,5)) 

in the following sample code:

<<Code, tidy=T, sanitize=TRUE>>= 
require(ggplot2)
df.qdf=data.frame(T     =     c(0    ,1,2  ,0    ,1.5,2.5)
                 ,q.500 =     c(0.025,2,1  ,0.025,  4,2  )
                 ,q.025 = 0.5*c(0.025,2,1  ,0.025,  4,2  )
                 ,q.975 = 1.5*c(0.025,2,1  ,0.025,  4,2  )
                 ,Occasion  = c(    1,1,1  ,2    ,  2,2  )
                 )
theme_set(theme_bw())
Graph=(qplot(T,q.500,data=df.qdf) 
      +geom_smooth(aes(ymin=q.025,ymax=q.975)
                  ,data=df.qdf
                  ,stat='identity'
                  ,fill='blue') 
      +ylab('Response')
      +xlab('Time [h]')
      +facet_grid(facet=.~Occasion)
      ) +scale_y_log10(ylim=c(0.025,5))
@

<<printgraph,echo=FALSE,fig=TRUE,width=10,height=7,sanitize=TRUE>>=
print(Graph)
@

under

• openSUSE 11.2,
• TeXlive2010, (update all installed as of today),
• R version 2.12.1 (2010-12-16) svn rev 53855 (x86_64-unknown-linux-gnu) with update.packages() as of today,
• LyX 1.6.8 (reconfigured after TeXLive update),
• pgfSweave of 18 December 2010 as configured according to Yihui Xie's script ('http://gitorious.org/yihui/lyx-sweave/blobs/raw/master/lyx-sweave-config.R') (Thank you Yihui!)

The code works ok

1. when pasted in the R-command window of rkward.
2. in lyx when I comment out the +scale_y_log10(ylim=c(0.025,5))

I have tried already

sanitize=TRUE
options(device = function(...) {     .Call("R_GD_nullDevice", PACKAGE = "grDevices") })

and

dev.off() 

(suggested in http://www.inside-r.org/questions/problems-ggplot-and-pgfsweave)

None of this helps. Please!

Answer 1

I must confess that this is directly my fault again (indirectly it might be Sweave to blame). I was too bold and optimistic on the undocumented ``null graphics device'' in R, and Cameron was mislead by me:

http://yihui.name/en/2010/12/a-special-graphics-device-in-r-the-null-device/

Although this null device can speed up Sweave a little bit and avoid producing unnecessary graphics files, it seems to be unable to handle all graphics systems. When I wrote that solution, I did not test carefully on ggplot2 (i.e. grid graphics).

> .Call("R_GD_nullDevice", PACKAGE = "grDevices")
NULL
> print(Graph)
Error in grid.Call.graphics("L_text", as.graphicsAnnot(x$label), x$x,  : 
  Metric information not available for this family/device
> dev.off()
null device 
          1 

For the time being, you can put this in the first code chunk of your Sweave document:

options(device = pdf)

This will bring up a redundant PDF file (Rplots.pdf), but can solve your problem anyway.

I will discuss with pgfSweave authors or Sweave authors later. This seems to be easy to improve in Sweave, but the problem has been there for many years.