Re: essential packages and Pre-Depends
Hi,
[If this discussion is restricted to essential packages,
please ignore this message. I am quite confused now]
>>"Santiago" == Santiago Vila <sanvila@unex.es> writes:
Santiago> -----BEGIN PGP SIGNED MESSAGE----- On 18 Feb 1998, James
Santiago> Troup wrote:
>> If procps is installed and is in an unconfigured state or even an
>> usable state temporarily (e.g. it's libc6 and libc5 hasn't been
>> unpacked yet), dpkg is not going to die, and in fact I suspect the
>> vast majority of {{post,pre}{inst,rm}} scripts won't.
Santiago> But it may happen. Remember the Murphy law.
Yes, but ... what is the minimal set that a pre and post depnd
may use? Should there be a policy restricting this for essential
packages? Failing any guidelines, it is easy to go too far and
maximize pre-depends. (are we just talking essential packages here?
in which case I withdraw my objection)
>> Even if they do, and even in the _exceedingly_ unlikely situation
>> that so many do die dpkg actually says "bugger it" and gives up,
>> the situation is not nice, but recoverable (simply install libc6 or
>> configure procps).
Santiago> So you are saying that "being able to arrive at a situation
Santiago> that is not nice" is *not* enough to use a Pre-Depends
Santiago> field?
Umm, well, where do we draw the line? I think i'd agree for
essential packages. But not for standard or optional packages.
Santiago> Are you saying that the (unknown, so far) harm caused by
Santiago> adding a Pre-Depends line is bigger than all those "not
Santiago> nice" situations you describe?
Well, a predepends line shall cause a break in processing of
packages to install if you use package ordering. By itself
pre-depends are not bad, but a large enough number may cause lots of
breaks in install (are we trying to minimize predepends?). For the
handfull of essential packages this is not too onerous.
Santiago> Could you explain in a few words which would be exactly, the
Santiago> harm caused by having Pre-Depends in all essential packages?
I don't know. There may not be any. I think there are
none. But I also think procps may well not be essential.
>> Thus even though procps is essential (IMHO it shouldn't be), if
>> it's unconfigured or even unusable, it's not going to make the
>> system FUBAR.
Santiago> So you think that Pre-Depends should be used only to avoid
Santiago> *completely* FUBAR states? What about half-FUBAR states?
Santiago> We should try to avoid *most* bad states, including those
Santiago> "not very nice" ones you refer. This is exactly why the
Santiago> package system was created.
There has to be restraint. I am just wondering where we draw
the line. We can't make all depends into pre-depends (except for
essential packages)
Santiago> I wonder: Why do you want to make the bo -> hamm upgrade
Santiago> such an unpleasant experience?
Santiago> We are already using all essential packages in pre,post
Santiago> scripts without declaring a dependence. Why do you want all
Santiago> those pre,post scripts to fail? Yes, you may say "I don't
Santiago> want them to fail, just that it would not be too bad if it
Santiago> happens". The question would be then:
Santiago> Why don't you want to *avoid* them to fail?
manoj
--
Miniscribe's troubles are daunting. The company has floundered in
its attempt to settle 13 shareholder lawsuits, filed after a panel
found that previous managers circumvented financial controls and
resorted to shipping bricks and unfinished drives to shore up sagging
revenue figures. "Miniscribe Prognosis Is Hopeful," E. E. Times, Jan
15, 1990, pg 67
Manoj Srivastava <srivasta@acm.org> <http://www.datasync.com/%7Esrivasta/>
Key C7261095 fingerprint = CB D9 F4 12 68 07 E4 05 CC 2D 27 12 1D F5 E8 6E
Reply to: