Skip to content

Problem with the update to R 4.0.2 in the Fedora users' list

4 messages · Iñaki Ucar, Tom Callaway, José Abílio Matos

Original
José Abílio Matos
# 
There was a thread this weekend in the fedora users' mailing list where a user 
had problems updating R 4.0.2:

"non-rpm R libraries not accessible now w R v 4.0.x"
https://lists.fedoraproject.org/archives/list/users at lists.fedoraproject.org/
thread/2FFST3GWZCNM45SX53VKB255TO4LOV4C/

TLDR; as far as I can see the user had installed (as root) a package from cran 
and had installed the same package (R-here) from the Fedora repositories.

Since the package were installed in different locations the package installed 
by the user appeared first and thus it won regarding the Fedora package. The 
only problem was that the other package was installed using R 3.6 and thus the 
user had the warning that seemed confusing.

Are there any kind of tools to pick these cases or this is one of those corner 
cases that are not worth the trouble?

Best regards,
Iñaki Ucar
#
On Tue, 21 Jul 2020 at 17:17, Jos? Ab?lio Matos <jamatos at fc.up.pt> wrote:
I don't follow that list, but I see you have everything under control, thanks.
IMO it's not worth the trouble. Users should never install packages as
root. That's it. This may sound harsh, but, if they do, then they
should know what they're doing, and it's their problem. This is like
the seals on the screws of an electronic device: we ship one
configuration that works; we simply cannot foresee the countless ways
to break it once you start tinkering with it. :)
Tom Callaway
# 
I'm inclined to agree. Users are expecting that things they install
manually will override the system default (if any).

Tom
On Tue, Jul 21, 2020, 11:39 AM I?aki Ucar <iucar at fedoraproject.org> wrote:

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    
2 days later

    

      
      

        


  

        

      José Abílio Matos
    

    

      
      #
    

  


  

      
On Tuesday, 21 July 2020 16.43.31 WEST Tom Callaway wrote:

            

      
      
      
    

        
FWIW I agree with both of you.

OTOH someone who never did that kind of error please raise your hand. :-)

Does anyone knows if there is some kind of packages repository lint that could 
catch this type of problems.


Those were the main points of this message.
What comes next it is just me thinking aloud since I am not proposing any, 
immediate or medium term, change this are just what I have been thinking for a 
long time.

One way to avoid this kind of errors would be to have the library directories 
with R version like I suggested in May. Searching for this on the web I see 
that users (mostly from windows and mac backgrounds) have some tips where the 
directories that R version used to installed them is present in the name.

The recipes for upgrading were basically to take the names from installed 
packages, save it into an external (csv) file and then with the new R version 
to install those packages. In another context I do the same thing when I need 
to install a new computer with Fedora and to have it working as soon as 
possible:

 in the previous computer
# rpm -qa --queryformat '%{name}\n' | sort | uniq > packages-list.txt

 in the new computer
# dnf install $(cat packages-list.txt)

And voil? a new computer is ready will all the same packages as the previous 
one.

I still think that this is a better naming scheme but looking briefly into the 
installed packages I see that perl and ruby do not follow it.

End of thinking aloud...