On Tue, Jun 23, 2020 at 11:02:02AM +0200, Michael Meskes wrote:
> Colin Watson wrote:
> > However, in order to make buster → bullseye upgrades work properly,
> > I think it's necessary to have bsdmainutils depend on bsdextrautils
> > for at least one release cycle.  Otherwise there may be a point
> > during the upgrade where col isn't installed and so man will be
> > broken; it's worth putting some effort into avoiding that because if
> > the upgrade happens to break then users may need to consult man
> > pages to work out what to do.  The only reliable way I can think of
> > to avoid this kind of problem is to have a hard dependency for a
> > while as a transitional measure.
> 
> Any idea how this scenario could unfold? I cannot imagine how it could
> get there. What I will do, though, is add a "Breaks: man-db (<<2.9.3-
> 1)" to bsdmainutils. Actually this is already in git.

Breaks only ensures that new bsdmainutils can't be unpacked until man-db
is deconfigured.  For example, it would still permit this plausible
upgrade ordering, which AFAIK apt would have no particular reason to
avoid:

 1. deconfigure old man-db
 2. unpack new bsdmainutils
 3. configure new bsdmainutils
 4. (piles of other stuff)
 5. unpack bsdextrautils
 6. unpack new man-db
 7. configure bsdextrautils
 8. configure man-db

man would be broken between the end of step 1 and the end of step 5.  I
think this is undesirable and unnecessary.

-- 
Colin Watson (he/him)                              [cjwat...@debian.org]

Reply via email to