Re: [dev] Opinions on GNU stow

From: Kamil Cholewiński <harry666t_AT_gmail.com>
Date: Wed, 30 Aug 2017 10:05:26 +0200

On Wed, 30 Aug 2017, Anselm R Garbe <garbeam_AT_gmail.com> wrote:
> Symlinks have always been a hack due to Unix' lack of a proper
> namespaces approach. Plan 9 later fixed this by introducting a proper
> namespaces approach[1] - but even today unices (incl. Linux) have
> almost ignored the learnings of Plan 9 with some exceptions.

You do have union filesystems and mount namespaces in Linux. Actual
businesses are running them in prod and betting billions on it.

> In terms of a packaging manager, I'm a proponent of the idea I
> introduced with stali as well. It does not require a package
> "manager", but uses git for the rootfs overlay instead. If you want a
> certain version of the system, you check out the required version from
> /.git.

This is excellent for the base system, but it leaves a lot of problems
unsolved, especially for managing optional/third-party software.

Having two separate package management strategies for base & everything
else is duplicated effort solving common problems, and added mental
overhead for both developers and users.

> I'm certain that those ideas would scale up to a general purpose base
> system, however if you want to deploy heavyweights like chrome or
> openoffice etc. I would try to adopt union mounting overlays into some
> /crap namespace of such "packaged" software, rather than using ugly
> symlinks with stow.

On my daily driver, which I'm trying to keep as lean as possible while
still getting actual work done:

> $ dpkg --get-selections | wc -l
> 1471

I'm not sure if your approach would scale well. Care to try 10's, 100's
of union mounts and check on overall performance and stability? At that
point we would need a tool for managing these mounts as well, and
something is telling me it would be awfully similar to stow (minus the
usual GNU bloat).

<3,K.
Received on Wed Aug 30 2017 - 10:05:26 CEST