On Mon, 21 Aug 2023, 02:35 Sam Hartman, <hartm...@debian.org> wrote:
> >>>>> "Adam" == Adam Borowski <kilob...@angband.pl> writes:
> Adam>all
> Adam> files from /lib /bin (that are not required interface, such as
> Adam> /bin/sh -- which can be symlinked) are moved within a week
> Adam> from the first upload. Rather than taking two+ Debian
> Adam> releases!
>
> This is an area where I have skepticism.
> It sounds like you and Ian believe this part is easy.
> My concern is that I think the number of interface points are larger
> than you expect.
> As an example, does pam look in /lib/security, or /usr/lib/security?
> Which switches first: the location of pam modules or the place the
> library looks in?
> How do we avoid systems in situations where logins break.
>
> If you can show a general answer that works for pam, systemd, and other
> that doesn't involve taking things on a package-by-package basis you
> would go a long way to convincing me to think hard about your approach.
>
> Even if you need to take things on a package-by-package basis, if you
> can show:
>
> 1) You have a good answer say for pam, systemd and nss
>
> 2) You know all the packages that will require special handling
>
> 3) Argue why you are confident in your answer to 2,
>
> then I would think much more about this proposal.
>
>
>
> >> If we carry out this transition package-by-package without
> >> central coordination ("the traditional Debian way"), it seems to
> >> me that the best we can achieve is for /bin, /sbin, /lib* to be
> >> symlink farms, consisting of symlinks that are either owned by
> >> the same package that owns the symlink target, or are unowned
> >> from dpkg's perspective and are created by maintainer scripts.
>
> Adam> That'd be the resulting state of the steps I mentioned above.
> Adam> Once /lib /bin (or preferably, /usr) are empty save for
> Adam> symlinks, they can be merged safely.
>
> Actually, it's not at all obvious to me why this is true.
> I think the work done for Dep 17 convinces me that even if /bin and /lib
> and /sbin were entirely symlinks, owned by multiple packages, doing the
> final merge is kind of tricky.
>
> I absolutely agree with Ian that this usrmerge mess is more complicated
> than some people believed back when debootstrap's default was changed.
> However, I kind of think it's complicated no matter which way you go,
> and I have yet to be convinced that moving files one package at a time
> works well or saves complexity.
> I think it moves the complexity around.
>
This is all evidence-free contrarianism and obstructionism, as per usual.
In fact, the only evidence we have proves the exact opposite: only one
distribution attempted this so called symlink farm approach that a couple
of people seem to love so much as an idea, and it failed miserably. Suse,
which has much more advanced tooling and infrastructure that we can only
dream of (for example rpm, which is able to track files by hashes wherever
they move, unlike dpkg which craps itself and falls on its face with its
pants down as soon as a symlink appears), and a more centralized governance
with few if any personal fiefdoms (cough), tried it for a few years, then
had to backtrack and go the way everyone else successfully went instead,
which quickly wrapped up as expected.
It's been almost 10 years, so certain people should either provide the
extraordinary evidence required to back these extraordinary claims, or just
get over it. Thanks to our work users and external developers can enjoy a
legacy-format-free Debian since bookworm, and thanks to Helmut's hard work
packagers will soon do too. It would be better to spend time helping him
rather than keep beating the very much dead symlink farm horse, I am quite
sure.
>
