Re: [PATCH] Add vcflib.

[Roel Janssen <roel@gnu.org>]

Dear reviewers,

Can I make the review process easier by posting the patches in separate
threads?

Thanks!

Kind regards,
Roel Janssen


Ricardo Wurmus writes:

> Roel Janssen <roel@gnu.org> writes:
>
>> Dear Guix,
>>
>> In an effort to package freebayes, I would first like to add vcflib and
>> its dependencies.  Therefore, I have attached eight patches.
>>
>> I've attempted to decouple the dependencies from vcflib, which worked
>> for intervaltree and htslib.  The following dependencies are actually
>> tightly coupled in the build process:
>> - tabixpp
>> - smithwaterman
>> - multichoose
>> - fsom
>> - filevercmp
>> - fastahack
>
> Thank you for your efforts!
>
>> To decouple these, we would need to include header files in the package
>> output that aren't part of the public interface.  In addition to that,
>> we would need to patch the build system to not look for <object>.o, but
>> instead add the right directives to the linker.
>
> I see what you mean.  This is ugly, but there isn’t much you can do
> about this.  I’ve encountered many packages that do the same (some of
> which also involve tabix), and in the case of “libvcflib.a” it really
> does seem very hard to untangle dependencies further.
>
> “filevercmp” and “fsom” don’t install libraries at all and apparently
> are supposed to be included in other code.
>
>> I don't think that is the desirable approach, because that would cause
>> these packages to provide header files that should only be used
>> internally.  Therefore, I use the source of these packages in vcflib,
>> and unpack them in the vcflib project root, to avoid confusion on
>> interfaces and fiddling with the build system.
>
> This seems acceptable to me.  I haven’t yet looked at your patches, but
> the justification you give is valid; I looked at the Makefiles of vcflib
> and included projects and I think this is okay.
>
> ~~ Ricardo