[R] Rcmdr X11 protocol error message

[John Fox]

Dear Michael and Peter,

I've made the following changes to the current development version of the
Rcmdr package (on my web site, at
http://socserv.socsci.mcmaster.ca/jfox/Misc/Rcmdr/index.html, but not posted
to CRAN):

(1) The default for the Rcmdr grab.focus option is now set to FALSE for
non-Windows systems.

(2) There's a new report.X11.warnings option, which is set to FALSE by
default; this suppresses these warnings so that they don't appear in Rcmdr
warning dialogs. This just papers over the problem without really solving
it.

I'm not happy with these "solutions," but perhaps they'll help until we
discover the source of the problems. (I tested on my Quantian system.)

Thanks for your help.
 John

> -----Original Message-----
> From: Michael Bibo [mailto:mbibo at qldnet.com.au] 
> Sent: Wednesday, September 01, 2004 6:21 AM
> To: John Fox
> Cc: 'Peter Dalgaard'; r-help at stat.math.ethz.ch; 'Dirk Eddelbuettel'
> Subject: Re: [R] Rcmdr X11 protocol error message
> 
> John Fox wrote:
> 
> >Dear Peter and Michael,
> >
> >I installed Quantian on  a spare machine that I have and 
> observed the 
> >same warning messages that Michael has been reporting. These 
> (and the 
> >problem with help files but not with viewing data sets that Peter 
> >reported) occurred with version 0.9-11 of Rcmdr but not with 
> an earlier version.
> >
> >Since the code for the Rcmdr package was substantially reworked this 
> >summer, that seems to me a good candidate for the source of these 
> >problems, though I don't see why the changes should be 
> problematic. I'm 
> >afraid that I'm insufficiently familiar with the inner 
> workings of X11 
> >and Tcl/Tk to be much help in figuring out what's wrong. Everything 
> >seems to work fine under Windows, as far as I can tell.
> >
> >It occurs to me that if the warning messages are benign, one 
> approach 
> >would be to suppress them. I already intercept warnings and present 
> >them in dialog boxes; I could grep for "X11 protocol error" 
> and simply 
> >ignore these. That doesn't seem to me a good solution, however. It 
> >would be better to understand what's happening.
> >
> >I'm copying this message to Dirk since he's mentioned that 
> he plans to 
> >put the newer Rcmdr in Quantian. Dirk: Have you tested with 
> Rcmdr 0.9-11?
> >
> >Thank you.
> > John
> >  
> >
> >  
> >
> John, Peter and Dirk,
> 
> I'm glad I'm not the only one with a handle on it now.  I've 
> already exceeded my level of knowledge and skill in both 
> Linux and R.  But I'm happy to help track it down, if I can 
> be of assistance.  As I have said to John, the major value of 
> Rcmdr to me is to help 'sell' R to others in my organisation 
> currently using SPSS, and for this purpose, the Windows 
> version at work is working fine.  It is more of an annoyance 
> at home, as it doesn't seem to stop anything working.
> 
> Peter asked:
> 
> BTW, sometimes Tk errors allow you to see a trace of the execution.
> Would this happen to be one of those situations?
> 
> The short answer is I don't know (see my first line :-)).  
> There is no button on the dialogue box saying 'more 
> details...' or anything that obvious.  Is there a log file 
> somewhere on the system or some other way to generate such a trace?
> 
> I have been experimenting, and the following seem reliable 
> (at least on my system):
> 
> Error messages don't seem to happen for the first graph 
> generated.  The second graph drawn also generates an error 
> dialogue box with the error message repeated about 11 times.  
> The third and subsequent times exactly the same graph is 
> generated leads to an error message repeated about 21 times.  
> (I don't know if the number of repetitions is meaningful).  
> Note that the graphs are generated and visible when the error 
> messages appear.
> 
> Running any analysis that only writes output to the output 
> window (such as fitting a regression model) generates the 
> error messages before the output is written to the output 
> window.  The output appears when the error dialogue box is 
> OK'd.  I guess it's more likely to be responding to exiting 
> the dialogue box than writing the output.  But this only 
> happens when such error messages have been generated 
> previously in that session by creating a (second) graph.
> 
> If the diagnostic panel plots for a regression model are 
> called, the error messages appear, but this time the plots 
> themselves are not visible in the graphics device (it is 
> blank).  When the error messages are OK'd, the plots appear 
> in the device.
> 
> When exiting Rcmdr, then, and OK'ing the "Exit?" dialogue 
> box, the final error messages appear, but again only if they 
> have already been generated by a graph call.
> 
> I don't know if this will be helpful, but I thought reliable 
> observations might give some clues.
> 
> Regards,
> 
> Michael
> 
> michael_bibo at health.qld.gov.au
> 
> 
>