Ludovic Courtès <ludo@gnu.org> writes:

>> Is there a better way to make sure that all (direct and transient)
>> runtime dependencies can be found through the PERL5LIB variable?
>
> Good question.  The #:inputs parameter of build phases contains all the
> inputs (native, normal, and propagated) when not cross-compiling, so
> that doesn’t help.
>
> Currently you would have to enumerate the subset of the inputs that you
> want to use in ‘wrap-program’ as done in the ‘clusterssh’ recipe, for
> instance.  That’s tedious but still preferable, esp. if that reduces the
> size of the package’s closure.
>
>> From a7f69eb16e91ca94e5894b234a98a7f14e78fd64 Mon Sep 17 00:00:00 2001
>> From: Ricardo Wurmus <ricardo.wurmus@mdc-berlin.de>
>> Date: Wed, 3 Jun 2015 17:44:20 +0200
>> Subject: [PATCH] gnu: Add BioPerl.
>>
>> * gnu/packages/bioinformatics.scm (bioperl-minimal): New variable.
>
> [...]
>
>> +            ;; Make sure all executables in "bin" find the required Perl
>> +            ;; modules at runtime.
>> +            (let* ((out (assoc-ref outputs "out"))
>> +                   (bin (string-append out "/bin/"))
>> +                   (path (string-append out "/lib/perl5/site_perl:"
>> +                                        (getenv "PERL5LIB"))))
>> +              (for-each (lambda (file)
>> +                          (wrap-program file
>> +                            `("PERL5LIB" ":" prefix (,path))))
>> +                        (find-files bin "\\.pl$"))
>
> So either leave a “FIXME” saying that this is closing over a superset of
> what’s actually needed at run time, or enumerate the inputs.
>
> Otherwise LGTM.

Attached is a new patch in which only the paths of the transitive target
inputs (+ the output perl site_perl directory) are wrapped around the
executables.  The resulting path is a lot shorter than the PERL5LIB path
at build time and judging from my quick tests the executables do not
complain about missing Perl modules.

WDYT?