Re: Can /usr/share/doc/<pkg> be deleted on upgrade ?
On Sunday 29 November 2009 04:53:05 Manoj Srivastava wrote:
> On Sat, Nov 28 2009, Jesús M. Navarro wrote:
> > Hi, Ben:
> > On Saturday 28 November 2009 08:59:13 Ben Finney wrote:
> >> "Jesús M. Navarro" <email@example.com> writes:
> >> > Not personal but sysadmin related. When I want to find information
> >> > about a given package I go to /usr/share/doc/<pkg> so I find
> >> > reasonable that the local sysadmin would add notes about the package
> >> > right there if needed.
> >> No, I don't think that's reasonable. The ‘/usr’ hierarchy (with the
> >> important exception of ‘/usr/local’) should be considered entirely the
> >> province of the package management system; any files there can appear or
> >> disappear as dictated by the packages.
> >> The sysadmin's site-local files should be going under ‘/usr/local’,
> >> which *is* out of bounds for the package manager.
> > Strongly questionable: notes about package emacs, installed via package
> > manager might go under /usr/share/doc/emacs, why not.
> Why not? Because it is not safe, that's why. There is no
> guarantee made by Debian that your files shall not be stomped on, or
> that user data will be preserved.
My general stanza is not that would be the best/more sensible place to put
files on (I already said I never did it) nor that there isn't a note on some
(quite hidden) place saying that's against procedure. It's a bit on a higher
level: there's some obvious common behaviour deleting whole (non-empty)
directories is not the usual way so unless there are strong reasons (it makes
a bit easier going that way because so does the upstream maintainer I don't
think qualifies) I'd favour *not* to do that. Do you really find intuitive
going to /usr/local/share/doc/emacs to look for extra docs about the
package-managed installed emacs -specially when this package already has all
its documentation under /usr/share/doc/emacs? Well, I don't.
> But Debian also does not tell you that your file will be there
> with the next upload. If you name your file foo.txt, there is nothing
> that guarantees that the next version will not have an empty file
> called foo.txt in that dir in /usr. Nothing checks to see i there is a
> user file there. And, by the same token, when the next+1 version
> removes foo.txt, dpkg will happily remove it.
True. But again by comparation to other similar behaviours I'd find quite odd
that the system would remove/replace /usr/share/<pkg>/local-notes.txt or
even /usr/share/doc/<pkg>/mycompany-notes.txt (I think I remember the
prefix "local-" to be safe at least under /etc). I know that only files
under /etc (well, files marked as "config") are safe to be tweaked by the
local administrator on Debian but even that shows more of a
limitation/compromise from the used tools than a real common-sense/best world
policy (it'd be better to track *all* files, i.e. by means of md5sums, were
it not too expensive/burdensome).
> So, the user is well advised not to trust any user data under
> /usr/share, should be using /usr/local anyway. Given that, while a
> trifle odd, I see nothing wrong in removing and recreating
> /usr/share/doc/<pkg> with every install.
Then why /etc/<pkg>, /var/lib/<pkg> or /var/log/<pkg> won't get the same fate?
What's *so* different about /usr/share/<pkg> as to expect it to be managed so
differently? (again, I am not saying that there isn't a reason nor that it is
even written down somewhere but I'm questioning the overall sensibleness of
such a policy; after all the main reason d'etre for a distribution effort is
giving a focused common behaviour and integration of an otherwise disparaged
bunch of packages; the less details/exemptions for the policy, the better).