[R] Latest revision of lme4 requires 'Matrix' >= 0.995.2 but that version is not available on CRAN

[Uwe Ligges]

Peter Dalgaard wrote:
> Søren Højsgaard <Soren.Hojsgaard at agrsci.dk> writes:
> 
> 
>>I just updated the packages on my pc (windows xp). When loading lme4 I get
>> 
>> > library(lme4)
>>Error: package 'Matrix' 0.995-1 was found, but >= 0.995.2 is required by 'lme4'
>>
>>'Matrix'  0.995-1 is indeed installed on my computer, but
>>update.packages() does not capture a never version; and seemingly
>>for good reasons: When looking at CRAN, the new version of Matrix is
>>available - but only as a .tar.gz archive. 
> 
> 
> Oh s.....
> 
> 
>>Two questions
>> 
>>1) How can a new version of lme4 make its way to CRAN when the packages it depends on are not available (that is, in a zip-version)?
> 
> 
> Looks like a buglet in Uwe's scripts. He has a check error on Matrix
> 0.995-2, but presumably it was actually built and loaded by lme4
> 0.995-2...
> 
> And of course the previous version of lme4 goes into the bit bucket
> immediately. 
> 
> lme4 0.995-1 file still exists on the DK mirror. I've put a copy for
> you in the usual place, just in case.
> 
> 
>   
> 
>>An answer could be: Buhh, you can just download the .tar.gz archive and then compile it yourself! Surely, that would be possible, but it can not be the intention...
>> 
>>2) When actually trying to compile Matrix, I get
>> 
>>lmer.o(.text+0x285f):lmer.c: undefined reference to `atanh'
>>make[3]: *** [Matrix.dll] Error 1
>>make[2]: *** [srcDynlib] Error 2
>>make[1]: *** [all] Error 2
>>make: *** [pkg-Matrix] Error 2
>>*** Installation of Matrix failed ***
>>Removing 'c:/programs/R/current/library/Matrix'
>>Restoring previous 'c:/programs/R/current/library/Matrix'
>> 
>>Can anyone help me on what to do???
> 
> 
> This is not the same error that Uwe had (that was a clean compile but
> error in the check phase), so upgrading to the latest version of the
> build tools could be required.
> 

We (Doug, Martin and I) are trying hard to debug an error in Matrix. 
Martin and Doug have ideas for a bugfix and I am checking right now.

The latest version of Matrix that passed the checks was Matrix_0.995-1 
which is still available from 
http://cran.r-project.org/bin/windows/contrib/2.2/last/

The error was also present in Matrix_0.995-1, but in that case (and by 
chance) it did not cause a crash during the examples or tests (what 
happens for Matrix_0.995-2 under Windows).

Uwe Ligges