[R] Does R always insist on sending plot output to a file?

[Matt Shotwell]

Donald, 

I was able to 'trick' R into writing plot data to a GNU Linux fifo. I
had forgotten that the fifo will block until there is a process at
either end (a writer and a reader):

At one terminal, create a fifo and set a program to catch output

$ mkfifo Rfifo
$ cat Rfifo

At a second terminal

$ R
> postscript(file="Rfifo")
> plot(0)
> dev.off()

-Matt


On Wed, 2010-08-18 at 23:21 -0400, Matt Shotwell wrote:
> Donald,
> 
> At least for the PDF device (I know you asked about png, but I believe
> they are similar), the answer no. Ultimately, this device calls the
> standard C function fopen, and writes its data to the resulting file
> stream.
> 
> If you're using GNU Linux, you might trick R into writing to a fifo (a
> named pipe, see 'man fifo'), or some other in-memory device, and read
> from it with another program. My initial experiments with this, however,
> were not successful.
> 
> A better solution here, would be to have the various graphics devices
> write to an R connection, as do most other R functions that input and
> output data. In this way, we could write graphics data to a RAW
> connection (rawConnection()), which is essentially a memory buffer. 
> 
> There are two obvious barriers to this:
> 1. C level I/O routines (e.g. fprintf) are heavily integrated into the
> graphics device code. Hence, accommodating R connections would require
> significant changes.
> 2. The graphics devices are mostly implemented in C, and there is (at
> present) no interface to R connections at the C level.
> 
> -Matt
> 
> On Wed, 2010-08-18 at 21:49 -0400, Donald Paul Winston wrote:
> > I need to write the output of a R plot to a Java OutputStream. It looks like
> > R insists on sending it's output to a file. Is there anyway to get bytes
> > directly from the output of a plot so I can write it with Java? Writing it
> > to a file is too slow.
> > 
> > Is there a parameter in the graphics device function png(..) that directs
> > output to a variable in memory? 
> > 
> > x <- plot(.)  would make sense.
> 

-- 
Matthew S. Shotwell
Graduate Student 
Division of Biostatistics and Epidemiology
Medical University of South Carolina