[R] Are there better ways to save and restore par() settings

[Duncan Murdoch]

On 14/05/2010 6:19 PM, David Winsemius wrote:
> On May 14, 2010, at 5:45 PM, Duncan Murdoch wrote:
>
>   
>> On 14/05/2010 3:51 PM, David Winsemius wrote:
>>     
>>> Some of the help page examples use the form:
>>>
>>> opar <- par(<something>)
>>>     .....plotting activities...
>>> par(opar)
>>>
>>> This seems to "work" well, yet I have read in some places
>>>       
>> Please include commented, retrievalbe citations for "some places".
>>     
>
> In the examples from help(par) I see:
> op <- par(mfrow = c(2, 2),
> # 2 x 2 pictures on one plot pty = "s")
>   # square plotting region,
>   # independent of device size
>   ## At end of plotting, reset to previous settings:
> par(op)
> # Alternatively, op <- par(no.readonly = TRUE)
> # the whole list of settable par's.
> ## do lots of plotting and par(.) calls, then reset: par(op)
> ## Note this is not in general good practice
> Was it only with the use of the "no.read.only" parameter that one is  
> supposed to avoid the form I illustrated. I had assumed it applied to  
> both alternatives.
>   

No, it applies to everything.  The general rule of "etiquette" is that 
if you change the par() settings, you should restore them when you're done.

>>> that it is  not the preferred method to keep you parameters from  
>>> getting  corrupted. What is the preferred method?
>>>
>>>
>>>       
>> The above, with the restore wrapped in on.exit(), as your sources  
>> should have said.
>>     
>
> Perhaps it was implied in a manner to which I was not properly  
> receptive. The only place I see on.exit() on the par help page is in  
> the final example. I do not see a link to that function in the that  
> help page, and I had read it but assumed it was related to "exit"-ing  
> an R session. On visiting the on.exit help page for the first time in  
> my life, I see that its primary use is with graphics device calls. So  
> we are being advised to write that wrapper function each time we might  
> be making a par-changing plot?
>
>   
Code contained in "on.exit" expressions is executed when the current 
function exits.  I would guess the first example didn't use this, 
because there's a fiction when running examples that they are run by the 
user at the command line:  so on.exit() would never execute.  (I don't 
know what happens if you put a naked on.exit() in example code; I would 
guess it would cause trouble, because it would interfere with the way R 
already handles the exit from running an example.) You need to read the 
comment, which says "At end of plotting, reset to previous settings:".  
The on.exit() wrapper is a way of ensuring that this happens when the 
code is in a function.  No matter how the function exits (even if it 
exits due to an error), R will attempt to run that code and restore the 
settings.

So it is safe to run the example code in almost any context, because it 
sets a par() value, then restores it.  But real code usually has 
something in between those two steps, and it may quit, or return from a 
function, or die due to an error.  The advice is to write things in a 
way that ensures that no matter what goes wrong, if you've changed the 
parameters, then you will change them back.

Does that make sense?

Duncan Murdoch