[R-sig-hpc] Rmpi in "Depends"

[Ramon Diaz-Uriarte]

Dear Brian,

Thanks. Yes, listing as "Suggests" is one approach (what I am doing
now, and what some of my other packages in CRAN that use Rmpi do), but
it seems somewhat like cheating (the fact is that the package does
depend on Rmpi) and not very elegant (I'd rather have the failure at
install time than at run time). So something like having the
dependency check without actually starting the LAM/MPI universe.

Best,

R.





On Thu, Jul 16, 2009 at 3:29 PM, Brian G. Peterson<brian at braverock.com> wrote:
> Ramon Diaz-Uriarte wrote:
>>
>> Dear All,
>>
>> I have one package where I listed Rmpi in the "Depends" line in the
>> DESCRIPTION file. But this leads to unwanted behavior: installing the
>> package will try to start the lam universe (I am using LAM/MPI here).
>> This will not work if done as root, and is annoying when done as a
>> regular user.
>>
>> I just checked, and no package in CRAN is listed that has a reverse
>> Depends on Rmpi. I've googled around, but haven't been able to locate
>> anything, though I think I remember this issue has appeared in the
>> past in some list.
>>
>> Any ideas of the best way to proceed?
>>
>> Thanks,
>>
>> R.
>>
>>
>
> Sure, list it as "Suggests" in the package DESCRIPTION.
> Inside your code, check to see if it is installed, and throw a warning or
> stop() error if the package is not available without starting up the lam
> universe.  Make clear in your documentation why you've done this.
>
> Regards,
>
>  - Brian
>
> --
> Brian G. Peterson
> http://braverock.com/brian/
> Ph: 773-459-4973
> IM: bgpbraverock
>
>



-- 
Ramon Diaz-Uriarte
Structural Biology and Biocomputing Programme
Spanish National Cancer Centre (CNIO)
http://ligarto.org/rdiaz
Phone: +34-91-732-8000 ext. 3019