Skip to content

grid.cap() requires more time?

6 messages · Baptiste Auguie, Simon Urbanek

Original
Baptiste Auguie
# 
Dear all,

I am puzzled by the following behavior of the new grid.cap() function,
which appears to run out of time when capturing the output of a
graphic. It works fine if I introduce a Sys.sleep(1) before executing
more code,

library(grid)

quartz()
grid.circle(gp=gpar(fill="black"))
gg <- grid.cap()
dev.new()
grid.raster(gg) ## completely blank
gg[gg!="white"] ## indeed

quartz()
grid.circle(gp=gpar(fill="black"))
Sys.sleep(1)
gg <- grid.cap()
dev.new()
grid.raster(gg) ## OK
gg[gg!="white"]

I tried to see if the problem was limited to the quartz() device but
for some reason the x11() device is not working for me in this R
version,

capabilities(what = NULL)
    jpeg      png     tiff    tcltk      X11     aqua http/ftp
sockets   libxml     fifo   cledit    iconv      NLS  profmem    cairo
    TRUE     TRUE     TRUE     TRUE    FALSE     TRUE     TRUE
TRUE     TRUE     TRUE     TRUE     TRUE     TRUE     TRUE     TRUE
Warning message:
In doTryCatch(return(expr), name, parentenv, handler) :
  unable to load shared library
'/Library/Frameworks/R.framework/Resources/modules/i386/R_X11.so':
  dlopen(/Library/Frameworks/R.framework/Resources/modules/i386/R_X11.so,
6): Library not loaded: /usr/X11/lib/libpng12.0.dylib
  Referenced from:
/Library/Frameworks/R.framework/Resources/modules/i386/R_X11.so
  Reason: Incompatible library version: R_X11.so requires version
42.0.0 or later, but libpng12.0.dylib provides version 36.0.0

sessionInfo()
R version 2.11.0 RC (2010-04-16 r51754)
i386-apple-darwin9.8.0

locale:
[1] en_GB.UTF-8/en_GB.UTF-8/C/C/en_GB.UTF-8/en_GB.UTF-8

attached base packages:
[1] grid      stats     graphics  grDevices utils     datasets
methods   base

loaded via a namespace (and not attached):
[1] tools_2.11.0

I would appreciate if someone could confirm this behavior. Pointers to
a fix for the x11() device on my machine are also welcome!

Best regards,

baptiste
Baptiste Auguie
# 
I just figured out what the strange "noise" consisted of in the
captured output of my previous example, and this is another source of
curiosity for me,

quartz(width=0.1, height=0.1)
gg <- grid.cap()
dev.new()
grid.raster(gg)

Should grid.cap() really capture the resizing handle of the quartz() window?

Best regards,

baptiste
On 17 April 2010 12:34, baptiste auguie <baptiste.auguie at googlemail.com> wrote:
Simon Urbanek
# 
Baptiste,

first, there is a mailing list specifically for Mac questions - R-SIG-Mac.

Now to your post - grid.cap captures the screen of the device which has two implications here:

a) Quartz is asynchronous -- i.e. is doesn't actually draw the content on screen until you're finished with drawing (for efficiency). Unfortunately R has no provision to tell the graphics device that it's done with drawing (you can always add another lines() etc.) so Quartz is simply guessing by measuring the time between draw commands. So when you run grid.cap while the output has not been drawn yet (i.e. immediately after the last drawing command) it will be empty as there is no content yet since Quartz is waiting for the end.

b) it will contain the resize mark because it is a screen shot so the mark is actually part of it (this is intentional).

If what you really want is a bitmap from the device, it's better to use quartz.save instead (followed by readPNG if you want the bitmap as raster) -- that actually re-runs the plot in a separate quartz device that is not on-screen so neither of the above are an issue. 

That said, you can file a) as a bug against Mac version of R (at https://bugs.r-project.org/ ) since grid.cap should actually trigger the flush before it does the capture. I cannot promise that the fix will make it to 2.11.0, though, because it may be non-trivial to trigger the asynchronous flush and wait for it without blocking something (I'll have to look).

Thanks,
Simon
On Apr 17, 2010, at 6:34 AM, baptiste auguie wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Baptiste Auguie
    

    

      
      #
    

  


  

      
Hi,

        
On 17 April 2010 18:51, Simon Urbanek <simon.urbanek at r-project.org> wrote:

            

      
      
      
    

        
I wasn't sure if it was Mac-specific (OK, quartz() is but I could not
test x11()).

            

      
      
      
    

        
OK, that makes sense.

            

      
      
      
    

        
This design decision surprises me. Would it be possible to have an
option not to capture this mark as well?

            

      
      
      
    

        
I thought the aim of grid.cap was to make it easier to capture a
bitmap copy (no need to create an external file). Is a screenshot more
useful?

            

      
      
      
    

        
Will do.

Thanks,

baptiste

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Simon Urbanek
    

    

      
      #
    

  


  

      
On Apr 17, 2010, at 2:43 PM, baptiste auguie wrote:

        

            

      
      
      
    

        
I don't think so because the mark is part of the Quartz view - it is not something the device adds so it is not a "design decision". Again I can only repeat that if you want a re-play of the draw commands you should use that instead - it's a different task.

            

      
      
      
    

        
I didn't write grid.cap() so I don't know what the intention was, but "capturing a bitmap copy" is exactly the above (that is why it is called "capture" I suppose). What you are requesting is something different - creating a new bitmap using the same device settings and quartz.save does that. It would be trivial to add a parameter to quartz.save to return the bitmap directly instead of a file, but R did not have direct bitmap support so it was not requested so far.

Cheers,
Simon

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Baptiste Auguie
    

    

      
      #
    

  


  

      
Hi,

I submitted a bug report today.

        
On 17 April 2010 22:34, Simon Urbanek <simon.urbanek at r-project.org> wrote:

            

      
      
      
    

        
This would be very useful, I think. In fact, it sounds like it might
be possible to convert  a set of graphical commands directly into a
raster representation, without creating an intermediate file nor
opening an interactive device window. That would be awesome.

Thanks,

baptiste