[R] Hunting for snow...

[Duncan Murdoch]

On 7/9/2006 7:39 AM, Brian Lunergan wrote:
> Prof Brian Ripley wrote:
>> On Sat, 8 Jul 2006, Brian Lunergan wrote:
>>
>>> Evening folks:
>>>
>>> I did an install.views of finance and econometrics and between the two of
>>> them R reported 'FracSim', 'RDCOMClient', 'snow', 'VGAM', and 'segmented'
>>> as missing dependencies. Now, I've managed to hunt down what appear to be
>>> current zip file copies of all of the packages but 'snow'. Is there a
>>> windows edition of the package out there someplace that will get along 
>>> with
>>> R v2.3.1 and if so, how do I find it?
>>>
>>> Please and thanks...
>> PLEASE read the README:
>>
>> http://cran.r-project.org/bin/windows/contrib/2.3/@ReadMe
>>
>> and see the check summary at
>>
>> http://cran.r-project.org/bin/windows/contrib/checkSummaryWin.html
>>
>> snow runs distributed R tasks across hosts: it has three modes and
>> prefers PVM and MPI if packages rpvm or Rmpi are available (which they 
>> are not for Windows).  As a result only some of its functionality can be 
>> used on Windows, and anyone who is going to get involved at that level 
>> of programming will very easily install the package from the sources.
>>
> Let me see if I understand this correctly. Something in the finance task 
> view would seem to consider snow a dependency; however, the package in 
> question has dragged it along for the ride into Windows, crippling its 
> usefulness.
> 
> Problem and a question. Having checked the file on the UofT CRAN mirror 
> that lists dependencies for the available files snow only appears attached 
> to three of them, none of which are listed as part the finance view or 
> anything I already have on my system. What is R seeing these as a 
> dependency for then? Perhaps the view maintainers can suggest something? If 
> it is not available for windows and is only partially functional there can 
> it be safely ignored as a dependency?
> 
> One other puzzle resulting from this. The check summary suggested problems 
> with 'FracSim', 'RDCOMClient', 'VGAM', and 'segmented' as windows packages 
> but yet the versions I found seemed to install without complaint onto my 
> system. Three gave back the successful unpack msg. VGAM did not, but 
> several of the demo scripts ran without complaint so I am assuming it is 
> comfortable with v2.3.1 on a windows system. Later versions of the four 
> than were tested in the check summary, perhaps?

The check summary does more stringent tests than just installing the 
package.  It checks whether the examples will run, and, if the package 
author supplies some tests, checks that test results match expected results.

For example, in the case of FracSim, the error occurred when trying to 
run the fracsim.1d example.  Apparently the example gave no finite 
values so plot() couldn't set up a coordinate system.  (You can see this 
if you click on the "ERROR" link in the summary.)

Since this appears to be doing random number generation, it may just be 
that the seed was set to a particularly unlucky value.  But the code 
does assume that no such unlucky values exist...

Duncan Murdoch