Ludovic Courtès (2015-12-09 16:58 +0300) wrote:

> Alex Kost <alezost@gmail.com> skribis:
>
>> Ludovic Courtès (2015-12-08 02:00 +0300) wrote:
>>> Otherwise there’s a risk of fooling the GC, as in
>>> <http://bugs.gnu.org/20765>.
>>
>> Hm, OK.  I don't fully understand the issue though.  So *.el files of
>> 'emacs-w3m-1.4.538+0.20141022' will be compressed, and some of these
>> *.el.gz files contain store file names.  Do these files fool GC?
>
> Yes.  When a derivation build completes, the daemon scans all the files
> in the derivation outputs for /gnu/store/… patterns, and then records
> those patterns in the ‘References’ table of /var/guix/db/db.sqlite.
>
> This is what makes sure that if /gnu/store/A refers to /gnu/store/B, and
> if /gnu/store/A is live, then /gnu/store/B will not be GC’.d

Thank you for the descriptive explanation!  I confess I'm almost
ignorant in these internals.

>> If so, I think 'emacs-w3m' should be configured with
>> '--without-compress-install', right?
>
> Sounds like it!

Great, I have updated and pushed 'emacs-w3m' patch.

Now since it is clear that it is potentially problematic to have
compressed *.el files, I have changed this patch to handle only compiled
autoloads as the compressed ones are unlikely to appear ever.  OK?