[R-SIG-Mac] Make check error building R 2.6.1 under OS-X 10.4.11

[Stephen Shiboski]

Thanks Simon. I compiled both again last night on the same machine to  
make sure.

 From 2.5.1:

 > dbeta(0, 0.9, 2.2, ncp = rlnorm(100))
   [1] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf  
Inf Inf Inf
  [19] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf  
Inf Inf Inf
  [37] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf  
Inf Inf Inf
  [55] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf  
Inf Inf Inf
  [73] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf  
Inf Inf Inf
  [91] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf

 From 2.6.1:

 > dbeta(0, 0.9, 2.2, ncp = rlnorm(100))
   [1] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN  
NaN NaN NaN
  [19] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN  
NaN NaN NaN
  [37] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN  
NaN NaN NaN
  [55] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN  
NaN NaN NaN
  [73] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN  
NaN NaN NaN
  [91] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
Warning message:
In dnbeta(x, shape1, shape2, ncp, log) : NaNs produced


On Dec 16, 2007, at 6:21 AM, Simon Urbanek wrote:

>
> On Dec 15, 2007, at 10:47 PM, Stephen Shiboski wrote:
>
>> Thanks a lot. I noticed that there is a binary for gcc 4.2.1  
>> (gcc-42-5531-darwin8.tar.gz) on the R developer's page. If this is  
>> the same as what Apple provides, I can install this.
>>
>
> It is built from the same sources... (Apple doesn't provide gcc-4.2  
> binaries for Mac OS X 10.4)
>
>
>> I also noticed that the code for dbeta.c hadn't changed since  
>> 2.5.1. So, I just re-built 2.5.1 on the same machine and it makes  
>> & checks w/o errors!
>>
>
> Ah, oh .. that is indeed interesting ... and does that build return  
> all Inf for:
> dbeta(0, 0.9, 2.2, ncp = rlnorm(100))
>
> If so, then I shall have a closer look ...
>
> Thanks,
> Simon
>
>
>>
>> On Dec 15, 2007, at 5:32 PM, Simon Urbanek wrote:
>>
>>> On Dec 15, 2007, at 6:49 PM, Stephen Shiboski wrote:
>>>
>>>> Trying to build R 2.6.1 on a G5 imac running OS X 10.4.11. I am  
>>>> using
>>>> Xcode 2.4.1 and the gfortran (4.2.1) from r.research.att.com/ 
>>>> tools/.
>>>> I am using following the configuration:
>>>>
>>>> ./configure --with-aqua --with-blas='-framework vecLib' --with- 
>>>> lapack
>>>> --with-jpeglib --with-libpng
>>>>
>>>>
>>>> Builds fine, but make check fails as follows:
>>>>
>>>> running code in 'd-p-q-r-tests.R' ...make[3]: *** [d-p-q-r-
>>>> tests.Rout] Error 1
>>>> make[2]: *** [test-Specific] Error 2
>>>> make[1]: *** [test-all-basics] Error 1
>>>> make: *** [check] Error 2
>>>>
>>>> Here's the error:
>>>>
>>>> Error: dbeta(0, 0.9, 2.2, ncp = c(0, a)) == Inf is not all TRUE
>>>> In addition: Warning message:
>>>> In dnbeta(x, shape1, shape2, ncp, log) : NaNs produced
>>>> Execution halted
>>>>
>>>> I sucessfully built and checked 2.5.1 using the same set-up. The  
>>>> only
>>>> thing that changed was the update to OS X 10.4.11 from 10.4.10.
>>>>
>>>> It builds & checks fine on my Intel machine with the same Xcode and
>>>> gfortran.
>>>>
>>>
>>> The issue is known and appears to be a problem with a specific  
>>> version of Apple's compilers generating incorrect ppc code. It  
>>> goes away when you update to Apple's gcc 4.2 (or probably also  
>>> when you revert to older compilers, although I didn't test that  
>>> route specifically).
>>>
>>> However, I'm surprised that 2.5.1 would work *if* you had exactly  
>>> the same setup, because the dbeta source code didn't change since  
>>> 2.5.1 at all. Are you 100% sure that you had exactly the same  
>>> compiler versions as now?
>>>
>>> Cheers,
>>> Simon
>>>
>>
>>
>