bug#21949: git-2.6.3: 108M download, but 23M installed

[ludo@gnu.org (Ludovic Courtès)]

Florian Paul Schmidt <mista.tapas@gmx.net> skribis:

> I noticed this earlier and wonder if it's a bug or not:
>
> When downloading the git-2.6.3 substitute from hydra it is a hefty
> 108M download, but installed in the store it's just 23M.

The fact that the result is small is due to hard links (under
libexec/git-core), resulting from deduplication performed by the daemon:

--8<---------------cut here---------------start------------->8---
$ du -msc /gnu/store/8xk88mnx31b6i9jz0n7x7zm4xrnknj3b-git-2.6.3
23	/gnu/store/8xk88mnx31b6i9jz0n7x7zm4xrnknj3b-git-2.6.3
23	totalo
$ du -mscl /gnu/store/8xk88mnx31b6i9jz0n7x7zm4xrnknj3b-git-2.6.3
249	/gnu/store/8xk88mnx31b6i9jz0n7x7zm4xrnknj3b-git-2.6.3
249	totalo
--8<---------------cut here---------------end--------------->8---

The problem is that the Nix archive format (Nar) is dumb and doesn’t do
anything sensible with hard links.  So the actual archive is indeed that
big:

--8<---------------cut here---------------start------------->8---
$ guix archive --export /gnu/store/8xk88mnx31b6i9jz0n7x7zm4xrnknj3b-git-2.6.3 | wc -c
259217552
--8<---------------cut here---------------end--------------->8---

Hydra compresses it with bzip2, hence the 108 MiB.

Git is obviously a pathological case.  However, the issue is a more
global one that I think we should address using content-addressable
storage:

  https://lists.gnu.org/archive/html/guix-devel/2015-03/msg00666.html

Ludo’.