Re: Filesystem Hierarchy Standard 2.0 (fwd)
In article <[🔎] m0xVh2K-0004owC@chiark.greenend.org.uk> you wrote:
A well-stated proposal, Ian... but it gives me heartburn nonetheless.
: In the far future we can remove the compatibility links from
: base-files. This has to be considered carefully, because after those
: links are gone installing an old package will make `duplicate'
: instances of the moved directories.
I've been living through the filesystem reorganzation in HP-UX between 9.X
and 10.X for the last year or so at work. The 10.X revisions ship with a
"feature" called "compatibility links", which provides the set of symlinks
required to allow binaries compiled under 9.X to continue to function under
10.X... much as what you're proposing supports.
We curse them constantly.
I really wish we could have just had a flag day and cut from old to new.
I can't help but wonder if we can't do a better job of this in Debian. I
suspect that the answer has a lot to do with whether (after we release 2.0),
we again populate 'unstable' with a copy of 'stable' and work on updating it
in place, or whether we instead start with an empty directory and populate it
from scratch with packages rebuilt to adhere to a fresh set of packaging
standards. In a sense, treat a new packaging standards release as a new
platform we need to port to (which it is!), rather than as a trivial
evolutionary step. The libc6 experience so far certainly smells a lot like a
port to a new architecture... and all this "g-suffix" stuff with the libraries
is sure a pain to deal with on platforms that have never had libc5...
I haven't thought this all the way through, and don't have time to do so for
another week or three, but my experience is that when we try to soften
transitions like this (where "we" means the software community in general),
we generate just as much frustration as we avoid...
Bdale
Reply to: