[Nix-dev] Thoughts: What about using aufs instead of symlinks + env vars?

Michael Raskin 7c6f434c at mail.ru
Tue Jun 30 10:34:07 CEST 2009


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Marc Weber wrote:
> Thanks, I got you now. So at least before union mounting all those store
> paths there should be a test checking for conflicts and a big fat
> warning.

This is where the problem lies. I know that if I have managed to make
some package work and you have manage to make another package work, no
nontrivial checking is needed to tell whether they can be installed in
the same user environment - I can _always_ install them both, maybe
renaming executables in the process (cf. MySQL replace vs standalone
replace now installed as replace-literal).

> I still think that all those benefits could be worth giving it a try.

If we go that way we lose some of the main benefits of Nix - ensured
package isolation.

> Probably there are even more benefits:
> * Because there are no more store paths in the output its easier to verify
>   that -j4 builds craeted the same result as -j1 builds

If checking is all you want, you could build it as a fake fixed-output
with hash 00..0 .

> * there is a higher chance beeing ablte to reuse say gentoo ebuilds.
>   Doing so would a add tons of packages with not much effort.

Unfortunately, if it is ensured that we can add Gentoo packages without
any effort that also means that we cannot reliably get rid of any
problems with them..

What you propose becomes Gentoo on unionfs with chrooted builds. I had
an LFS system built approximately that way. Nix is much cleaner.

> So this is more an alternative which could work in most cases.
> It can't be a full replacement.

Yes, and we still need to fix the problems with "normal" Nix
installation. But if we fix them why not use the fixed package anyway?
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iQEcBAEBAgAGBQJKSc3/AAoJEE6tnN0aWvw3qbcIAIAcw0iPGaIAodsuuFZrHdSV
Q8Xv3us5etFrCo5f2+s5EScZKZyOLOW+ZwJS7WZDvP7J7D76UUx9bovKU6j3shu2
tLpRbmjdatMzQBZh+UZgePI4miVSqYqusoLU0SZ/tMMNpC2ckMzAN4HFSJzE2VCu
8I9guBbPFB94ykzQ4ToL/mp3x0MVyXQ1cRbzgpwmXdfFWkjqS6DCQMFlAu1u4uzZ
UJidYHZNPSGUjLc6b/mm46Qh+ECTdyeybAOkxgBGY13G/VN3dBJXFtZ5zhhoSnY0
Mh0pe92+7X4b35gFButxNOTpRQzzThhz/jAWkw8YJ9RoCqiWvYk2DpEeS9R6SSA=
=FUuA
-----END PGP SIGNATURE-----



More information about the nix-dev mailing list