Re: [PATCH] gnu: Add freebayes.

[Roel Janssen]

Leo Famulari writes:

> On Wed, Mar 09, 2016 at 07:53:20AM +0100, Pjotr Prins wrote:
>> On Tue, Mar 08, 2016 at 04:44:13PM +0100, Roel Janssen wrote:
>> > One of the problems with the patch is probably the bulk of dependencies
>> > dragged in (for example, vcflib).  They use specific versions so they
>> > are tied to this package (so that's why I cannot package them separately).
>> 
>> This approach is very common in bioinformatics and one way they fight
>> dependency hell. Not the best way, admittedly.
>> 
>> It may be worthwhile to package vcflib, for example, separately as
>> freebayes merely requires the header files to compile. The problem is
>> that that te current vcflib does not include the headers from a
>> default install, if I understand Roel correctly.  Ricardo, can you
>> advice on this? Should we add the headers in guix through a vcflib
>> installer?

I would like to propose the following:

- Separate the packages, and when needed, add specific versions of the
  packages.

  Instead of trying to package arbitrary Git commits, I will look into
  whether I can use release versions instead, without impacting the final
  FreeBayes binary.

- Use ,(package-source ...) to unpack the sources of the packages to
  satisfy the "submodules" structure of projects that use this.

This way we can separate the packages, which yields the programs that
are now hidden in submodule sources.  It also gives us more complete
package descriptions for them.  We don't have to heavily modify the
project setups that rely on submodules (just extract sources the way
we do now).

I am working on doing this for FreeBayes.  It should yield separate
packages for: vcflib, tabixpp, multichoose, smithwaterman, filevercmp
and fsom.

I want to find out whether I can make "intervaltree" a library, as it
is meant to be:
  https://github.com/ekg/intervaltree

WDYT?

Kind regards,
Roel Janssen