Bioconductor in Fedora

[Alex Lancaster]

This is just a heads-up about packaging of Bioconductor.  Currently
Pierre-Yves is spearheading the packaging of Bioconductor, and I'm
helping out with reviews.   Initially Pierre started with the unstable
(will be 2.2):

http://www.bioconductor.org/packages/2.2/bioc/

but that's a moving target and won't be officially released until 2008
and is targeting R 3.0 (?) so we dropped back to the stable (2.1)
release:

http://www.bioconductor.org/packages/2.1/bioc/

Unfortunately a couple of packages (R-tkWidgets and some others) from
2.2 have already made their way into Fedora, but they are only at most
1 or 2 minor releases away from 2.1, so in discussion with Tom
Callaway and Pierre-Yves on #fedora-devel, we decided to stick with
those packages but use only 2.1 releases for everything else.  Once
2.2 is released we can update all packages to that then.

I'm currently working on the review of R-Biobase, the core package:

https://bugzilla.redhat.com/show_bug.cgi?id=240500

We could perhaps also create a tracker bug and make all Bioconductor
module package reviews block the tracker bug.  If anybody would like
to help, let us know!

Cheers,
Alex

[Pierre-Yves Chibon]

Alex Lancaster wrote:
>  This is just a heads-up about packaging of Bioconductor.   
This last email was some time ago and the situation evolved a bit.

We have now more R libraries in Fedora and 15 libraries from 
Bioconductor. I have been talking about Bioconductor with Alex and we 
end with the idea that having at least the most important libraries of 
bioconductor available on Fedora would probably be a nice feature for 
bioinformaticians.

I have made a draft version of this feature :
https://fedoraproject.org/wiki/Features/Bioconductor

But I think some point should be discussed about bioconductor :
* What should we do about the metadata and the experiment data packages 
that are required (well in fact suggested most of the time)?
Those libraries contain only data. For the couple of package of this 
type that we have already in Fedora the solution until now has been to 
ask real-eng to inherit the tag between the version (from F-7 to devel). 
This process would solved quite some place on the server but have the 
problem that we have to annoy the real-eng for each package/update.

* Would it be possible to have a group package such as users just have 
to type /yum install bioconductor/ to get the basis package of 
bioconductor ?

* Is there a way to categorize the packages such as the user can easily 
found what packages from Bioconductor are available on the repo ?

* The list of packages to do in order to have the main packages of 
Bioconductor is (I expect, I have not checked yet) around 60 to 70.
I can maintain some but I don't want to do all myself, especially since 
my situation might change quite a bit in January.
Creating the specfile is not hard, here is a small python script that 
helps to generate the specfile [1]
Maintaining those packages is not really difficult, there is 1 release 
every 6 month (for each major release of R).
I developed a small python script [2] that handle most of the work, I 
still have some ideas to improve it (I want to really personalize it for 
Bioconductor packages).

So to conclude
* I think having Bioconductor in Fedora is a nice Feature
* I cannot make it on my own, but I have no problem to coordinate things 
in order that we build a group that could handle both the packaging and 
the maintaining of the packages.


Thanks to have read my bad English all the way to here :)
Let me know what you think of this,

Best regards,

Pierre


P.S. If you think that the project is nice, maybe we could look for more 
people through the devel mailing list ?

[1] http://pingoured.fr/public/R2spec_1.3.py
[2] http://pingoured.fr/public/updateCVS_0.2.py

[Pierre-Yves Chibon]

Pierre-Yves wrote:
>  Alex Lancaster wrote:
> > This is just a heads-up about packaging of Bioconductor.  
>  
>  This last email was some time ago and the situation evolved a bit. 
<snip />
>  * The list of packages to do in order to have the main packages of 
>  Bioconductor is (I expect, I have not checked yet) around 60 to 70.
>  I can maintain some but I don't want to do all myself, especially since 
>  my situation might change quite a bit in January.
>  Creating the specfile is not hard, here is a small python script that 
>  helps to generate the specfile [1]
>  Maintaining those packages is not really difficult, there is 1 release 
>  every 6 month (for each major release of R).
>  I developed a small python script [2] that handle most of the work, I 
>  still have some ideas to improve it (I want to really personalize it for 
>  Bioconductor packages). 

I have been looking a bit at the dependencies question.
There is 20 main libraries of Bioconductor  (based on what the biocLite 
script install by default). The list is presented here
https://fedoraproject.org/wiki/Features/Bioconductor#Scope

With the help of Torbjorn on #R (irc.freenode.net) I modified a small 
perl script[1] that goes on the bioconductor SVN and for each packages 
retrieve both the depends and suggests dependencies.
The results is available here:
http://pingou.fedorapeople.org/Bioconductor/Dependencies2
To compare with
http://pingou.fedorapeople.org/Bioconductor/Dependencies
When there is only the Depends taken into account.

 From that list I used this script[2] to process the dependencies and 
sub-dependencies that we would need to package.
The results are presented here:
Depends+Suggests:
http://pingou.fedorapeople.org/Bioconductor/Dependencies-Depends_Suggests...
Depends only:
http://pingou.fedorapeople.org/Bioconductor/Depedencies-Depends.txt

The number of libraries to package jump from 36 to 75 if we take into 
account the Suggests dependencies.
This only takes into account the dependencies of the software part of 
Bioconductor, the libraries called 'metadata' or 'experiment data' which 
I believe can also have sub-dependencies are not presented here.

This can give us some insight on where to start.
However I would like to know what you, guys, think of the question 
dependencies "Depends" vs "Depends+Suggests".
"Suggests" add more work and leads to the problem of circular references 
quite often...

I have not heard from you about the global idea, do you 
like/appreciate/don't care about it ? (cross out the wrong option)

Regards,

Pierre


[1] http://pingou.fedorapeople.org/Bioconductor/bio.pl
[2] http://pingou.fedorapeople.org/Bioconductor/showDep_1.1.py