[bug#39807] [PATCH] guix: pack: Only wrap executable files.

[Ludovic Courtès]

Hi Eric,

Eric Bavier <address@hidden> skribis:

> On 06.03.2020 05:16, Ludovic Courtès wrote:

[...]

>>> I feel like a test should be added to
>>> tests/guix-pack-relocatable.sh, but
>>> I'm not sure how to do that while keeping the test lightweight.
>>> Suggestions
>>> welcome.
>>
>> Not sure how to do that.  Since ‘guix pack’ accepts manifests, you
>> could
>> have a manifest containing a ‘computed-file’ with a file that shouldn’t
>> be wrapped, and then you could ensure that’s indeed the case.  Or you
>> could try with ‘git-minimal’ or some other package that exhibits the
>> problem?
>
> I almost have a working test using 'git-minimal', but I'm not happy
> with the quantity of code needed to setup, and I'm worried now that
> that test would be relying on an implementation detail that could
> change in the future without us noticing (e.g. a git subcommand that's
> currently a shell script is subsumed into git so the test no longer
> checks what we want).
>
> So I think I'll try going the manifest/computed-file route instead.

OK.

>>> -          (for-each build-wrapper
>>> -                    (append (find-files (string-append input "/bin"))
>>> -                            (find-files (string-append input
>>> "/sbin"))
>>> -                            (find-files (string-append input
>>> "/libexec")))))))
>>> +          (receive (executables others)
>>
>> I’d prefer srfi-11 ‘let-values’.  :-)
>
> I tried let-values to begin with, but I found 'receive' to be much
> easier on the eyes.  For the case of binding values from a single
> expression, does let-values offer benefits?  And there are no other
> uses of let-values in this module, so precedent/consistency doesn't
> seem to have weight.

OK, no big deal.

There are probably more uses of ‘let-values’ than ‘receive’ overall.
That said, I think we can start switching to srfi-71, which is nicer
than both of these.

Thanks,
Ludo’.