[Rd] X11 help please

[Duncan Murdoch]

On 6/18/2007 7:56 AM, Hin-Tak Leung wrote:
> Prof Brian Ripley wrote:
>> rgl normally works under Xvfb on Linux: we've been using that for the 
>> testing 'for ever'.
>> 
>> However R should not 'crash' on an error in a package, and the X11 device 
>> has error handlers set up by
>> 
>>  	XSetErrorHandler(R_X11Err);
>>  	XSetIOErrorHandler(R_X11IOErr);
>> 
>> I don't see anything comparable in rgl.
>> 
>> The error seems to be in the attributes, so I would have a closer look at 
>> those.  In particular, how are you running Xvfb?: you may need something 
>> like
>> 
>> Xvfb :5 -screen 0 1280x1024x24 &
>> setenv DISPLAY :5
>> 
>> to ensure that you have a reasonable visual and hence colormap.
>> (On my system the default is 12 bit.)
> 
> That's interesting - my first thought of Duncan's problem was unusual
> colour depth as well, but I thought very few people run their X servers
> in anything other than 24-bit or true-color mode these days (and any
> MacOSX hardware should be capable of that anyway); if 12-bit is the
> Xvfb default it is hightly likely that's the reason.

I don't know the default, but I've been specifying 24 or 32 bit depth 
each time, and rgl is seeing 32 bit depth.

> I have also seen that sort of error messages in CJK related issues,
> if the X server's font path has some unsual and broken fonts; so it 
> might be a good idea to disable any non-english locale, etc and try not 
> to do anything clever in that direction for the time being.

I do get a warning about a missing font when I run Xvfb; Simon Urbanek 
thought it was ignorable, but maybe not:

Could not init font path element /usr/X11R6/lib/X11/fonts/CID/

This happens before R is running, and sometimes again during the run.

I've now implemented XSetErrorHandler(R_X11Err) as Brian suggested, and 
this moves the error elsewhere:  R doesn't quit when it hits that error, 
but it dies later because rgl isn't handling its incomplete 
initialization properly.  I think that will be relatively easy to track 
down and fix, but I'm still left with two problems I don't know how to 
solve:

1.  Why am I getting this error in the first place?  Since things work 
on Linux, this may be a bug in the MacOSX implementation of Xvfb, but 
Google doesn't turn up other reports of that, so it's more likely an rgl 
problem.

2.  XSetErrorHandler() is a global setting.  If rgl sets it, then it 
will override the setting that X11() used.  Is there a way to retrieve 
the previous setting, so I can set it only temporarily?  There doesn't 
appear to be any XGetErrorHandler() function.

Duncan Murdoch
> 
> - I do run my old laptop at 16-bit only, as it is old and I get better 
> performance that way with little sacrifice on visual appearance,
> but that's my choice...
> 
>> On Sun, 17 Jun 2007, Duncan Murdoch wrote:
>> 
>>> The rgl package currently crashes R when running under Xvfb (the
>>> "virtual frame buffer" server), at least on MacOSX.  It makes sense that
>>> it shouldn't be able to work there (it needs interactivity), but I don't
>>> know how to detect the problems before they cause the crash.
>>>
>>> Currently the error happens the first time you try to open an rgl
>>> window; when rgl calls XCreateWindow R crashes with the error message
>>>
>>> X Error of failed request:  BadMatch (invalid parameter attributes)
>>>  Major opcode of failed request:  1 (X_CreateWindow)
>>>  Serial number of failed request:  28
>>>  Current serial number in output stream:  29
>>>
>>> I've checked and as far as I can see all the parameters we're sending to
>>> XCreateWindow are valid, but I might have missed something:  I don't
>>> know much about X11 programming.  (The call is
>>>
>>>  ::Window xwindow = XCreateWindow(
>>>    xdisplay, RootWindow(xdisplay, DefaultScreen(xdisplay)),
>>>    0, 0, 256, 256, 0,
>>>    xvisualinfo->depth,
>>>    InputOutput,
>>>    xvisualinfo->visual,
>>>    valuemask,
>>>    &attrib
>>>  );
>>>
>>> but without context that's probably not worth much to others.)
>>>
>>>
>>> Are there any experienced X11 programmers out there who can suggest what
>>> to do next?
>>>
>>> Duncan Murdoch
>>>
>>> ______________________________________________
>>> R-devel at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>>
>>