[Rd] More strange R CMD build/check errors on Windows

[Uwe Ligges]

On 12.09.2010 12:10, Hervé Pagès wrote:
> Hi Peter,
>
> On 09/12/2010 01:51 AM, Peter Dalgaard wrote:
>> On 09/12/2010 08:10 AM, Hervé Pagès wrote:
>> ...
>>>
>>> AFAICT those problems were never seen before (i.e. with R< 2.12).
>>> They show up randomly everyday for a small number of packages
>>> (between 10 and 20 out of 400). The set of victims changes everyday
>>> and any package seems to be a potential victim (I've not been able
>>> to observe any obvious pattern so far).
>>> Does anyone have any idea what could make 'R CMD build' and
>>> 'R CMD check' so confused/unreliable on Windows?


Brian had some ideas that the problems are related to the shell that is 
used. Is the problem still apparent in a very recent R-devel from few 
days ago? I am just back from vacations and have not updated yet.
I experienced the same problems and I am just iterating automatically if 
typical problems are apparent from the log files. I hope some if not all 
parts are solved now and will do some new test runs shortly.

Best,
Uwe






>>> Thanks,
>>> H.
>>>
>>
>> Antivirus software? I suspect you already ruled that out, but it has
>> been the culprit for problems with mysteriously disappearing
>> intermediate files in several cases, so I thought I'd mention it.
>>
>
> Actually I didn't try that yet because we still build BioC release
> (using R-2.11.1) on these 2 Windows boxes and we don't see any of
> those problems for the release builds. But I will. Could it be that
> the fact that 'R CMD build' and 'R CMD check' are R-based in R-2.12
> (and not Perl-based anymore) make them more fragile when something
> like an antivirus is messing around with the filesystem?
>
> Thanks for the suggestion,
> H.
>