Volker A. Brandt wrote [06/08/2008 08:39]:
> Hello Stephen & Bill & List!
>
> It's great to see some real-life examples going through this list.
> Ideally, it would be very instructive if Bill could post some 
> "before - after" comparisons once he has converted his packages.
>   

I'd be happy to do that, as and when I get to that point.  At present 
I'm just trying to assess how much time it is likely to take to do the 
migration, and the result of that will determine when (or indeed 
whether) we go ahead and do it.  That is very much an open question at 
the moment: I already have pkgadd and RPM versions of the package to 
maintain, and I'm going to take some convincing that what I really 
really want is to add a third version to that list.

>>> - check for a couple of prerequisite user accounts: abort install if
>>> they are not found
>>>       
>>   Use user action to install user accounts; fails if privileges are
>>   insufficient or accounts don't exist.
>>     
> Maybe I misunderstand, but I think the original intention was to fail
> without creating a given account if it did not exist.  That nut is a
> little harder. :-)
>   

The "original intention" is not entirely fixed.  The existing pkgadd and 
RPM implementations of my package check for the existence of a couple of 
required user accounts, and abort if they are not found.  That logic 
lives in preinstall (for pkgadd) and %pre (for RPM).  The same behaviour 
would be acceptable in an IPS implementation, but a better alternative 
would be to actually create the user accounts and carry on, rather than 
aborting.  (Technically I could have done that in pkgadd/RPM, but it 
seemed a bit scary ...)

>>   We don't have support for smf(5) service suspension yet, but we will.
>>     

Maybe I shouldn't have used the term "service".  The package in question 
isn't a service, in the sense used by SMF; maybe it should be, but 
that's another change of functionality which we would need to assess.  
What I would really like is the ability for IPS to call an arbitrary 
command in the package itself.  That would be useful both at upgrade 
time and at uninstall time.

>>   Please feel free to describe the other behaviours; we'd be happy to
>>   look at the postinstall as well--much of it can probably be absorbed
>>   into actions or action attributes.
>>     
> I would very much like to see such a discussion.
>   

I really don't want to get into a discussion of postinstall behaviour.  
My postinstall script is a 3000+ line shell script, which calls numerous 
other scripts and executables - all of which are specific to the 
functionality of the package itself.  Even in the pkgadd/RPM 
implementations most of this is now decoupled from the actual package 
installation, and is attached instead to the first instance of the 
package's "start" command.

--
Bill
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to