Re: New build system: copy-build-system

[zimoun]

Hi Pierre,

On Sun, 26 Jan 2020 at 21:27, Pierre Neidhardt <address@hidden> wrote:

> Those packages end up being tedious to write because we need to rely on
> the trivial-build-system, also known as "the least trivial of them all!" :)

I agree!


> What about adding a "copy-build-system" which has one argument, say an
> `#:install-plan` that takes a list of source-destination pairs?

I agree that something is missing.


> This would drastically simplify the definition of the aforementioned
> packages, basically summing up dozens of boilerplate lines into an
> effectively trivial file mapping.

I agree too.


> If this sounds like a good idea, I'd be happy to send a patch!

Related to this idea, I would point to [1] a previous discussion about
"pure" data package distribution.

[1] https://lists.gnu.org/archive/html/guix-devel/2018-02/msg00183.html


For a concrete example, let consider the Debian package [2]
'astrometry-data-2mass-00' which is roughly 13.9GB of data. A big set
of index used by astronomer. The Debian package calls 'curl' under the
hood to fetch the data from the http://data.astrometry.net; see [3].

[2] https://packages.debian.org/fr/stretch/astrometry-data-2mass-00
[3] 
https://salsa.debian.org/debian-astro-team/astrometry-data-2mass/blob/master/debian/astrometry-data-2mass-00.postinst


Well, the questions are:
 - hash?
 - a copy on Berlin as substitute?

I am just taking the occasion to discussion the hypothetical
possibility to discuss if 'copy-build-system' could be way to
distribute such data set.

Other said, the 'copy-build-system' could:

 - fetch the data from an archive, such http://data.astrometry.net
 - fetch the resulting <hash> of /gnu/store/<hash>-name-version from susbtitutes


Cheers,
simon