Please see inline ...

On Tue, Aug 21, 2012 at 8:55 PM, Joerg Schilling
<[email protected]> wrote:
> Moinak Ghosh <[email protected]> wrote:
>
>>    It has been a long time, approx 5yrs since I last touched that code and
>>    difficult to recollect everything. Some tidbits I do remember:
>
> Thank you for your reply...
>
>
>>    1. SVR4 does not have the concept of package upgrade. So if for eg
>>        pathnames are changed in a new package release old one has to be
>>        uninstalled and new version reinstalled to do a clean update. However
>>        it is not possible to do this with core system packages, so it needs
>>        wrapper scripts or custom class-action scripts. So pathname changes
>>        in core OS pkgs may also require system config updates. There is no
>>        support for doing these other than through custom wrapper scripts.
>
> The data base has a lot of related concepts and the mportant thing to know is
> that you first need to install the new package and then remove the old one.
> There are reference counters and there is the fact that files are actually 
> only
> removed when they have not been changed since the install of a to-be-removed
> package. If there are problems with upgrades, I expect only small problems in
> special as the package system supports versioned package installs. What I need
> to check is what happens after MAXINST upgrades have been passed.
>
> What really is missing is the knowledge that a package is never than another
> one and support for versioned dependencies (although there is a parser for
> versioned dependencies already).
>
>
>>    2. The concept of action scripts is a powerful one but with no control
>>        it creates a complete mess. For eg every team in SUN that ever 
>> provided
>>        a driver had their slightly different variant of the driver
>> install action
>>        script. It was a mess of action scripts all around.
>
> This is just a result of bad management inside Sun.
>
> I was really shocked to see the mess here, but this is something we can dix 
> and
> I already started to write standard scripts to be used via include.
>
> The worst package here is SUNWcsd that should not include a static
> name_to_major table.
>
>
>>    3. I remember a team member fixing a customer escalated bug when he
>>        figured out that a particular supposedly binary search for looking up
>>        pathnames was actually doing a linear search - some messy code.
>
>>        This only points to the age of the codebase, 25yrs approx. Old and
>>        somewhat clunky with a whole bunch of commands and libs.
>
> Isn't this something that should be be fixed with pkgserv?
>
> I am able to install the whole ON base within less than 2 minutes over the
> network, even though the packages are compressed via xz -9. It would be of
> interest to know what time you get.....
>
> Could you please fetch:
>
>         
> http://download.berlios.de/pub/schillix-on/packages/pkgs-schillix-on.txt
>
> and follow the instructions. Note that this works directly with pkgadd over 
> the
> network.

   Will check ...

>
>>    4. The approach of keeping all pathnames in a single large text file
>>        (/var/sadm/install/contents) is suboptimal. While this was updated
>>        by Casper Dik sometime back there are other such things. Limitations
>>        had only been patched over and over without any significant cleanups
>>        or re-writes to fix old design limitations.
>
> See above, I cannot see and real problem with it today.
>
>
>>    SVR4 had excellent concepts, in fact was state of the art 25yrs back.
>>    The concepts are still awesome even by today's standards but
>>    implementation leaves much to be desired.
>
> The packages system was created in 1982, so it is 30 years old. But not the
> time of creation matters...

   Age will not matter if the code is properly maintained. How old is
the Solaris
   kernel ?  SVR4 code has been badly maintained due various factors including
   bad management that I do not like to discuss here.

>
>>    Adding missing pieces to SVR4 will mean first and foremost a repo
>>    mechanism, package management and seamless upgrade support.
>
> I need to add a base URL for automated network installs, but this does not 
> seem
> to be hard to do.
>
>>    Once again trying to re-do solved problems. I was forced into the
>>    unfortunate wheel re-invention for an earlier BeleniX version because
>>    we did not want to touch IPS with a barge pole and we had no choice
>>    in the short term but to develop Spkg:
>>    http://moinakg.wordpress.com/2008/11/22/the-belenix-package-manager/
>>
>>    
>> http://belenix.svn.sourceforge.net/viewvc/belenix/trunk/spec_files/ext-sources/spkg?revision=175&view=markup
>>    
>> http://belenix.svn.sourceforge.net/viewvc/belenix/trunk/spec_files/ext-sources/spkg_mod.py?revision=395&view=markup
>>    http://moinakg.wordpress.com/2008/11/22/the-belenix-package-manager/
>
> Thank you, I'll read this later....
>
>
>>    It had all the useful features and brought in at least one concept that
>>    "I dare say" was discussed within the IPS team later, without attribution
>>    of course. However it was never intended to be a full fledged replacement
>>    before people jump to find flaws in that.
>>    It took about 6 months part-time effort for a single person to develop
>>    compared to the 3-years of elaborate intricate complex wheel re-invention
>>    with the intention to solve world package hunger and even do, surprise
>>    surprise Windows package installation!
>
> I had some discussions with Dagobert Michelsen from opencsw already and he
> seems to have a lot of knowledge in this area on the svr4 package system. It
> seems that we just need to make the important decisions now and could 
> implement
> most of the code later.
>
>
>>    I want to give a wide berth to both IPS and the existing SVR4
>> "implementation".
>>    I actually respect SVR4 as a concept but am not interested in doing 
>> anything
>>    with its current codebase.
>
> Well, if we look at what existing OpenSolaris based distros use, there only
> seems to be IPS and from what I've seen with IPS during the past 2 years, I
> don't like to use it.

   The worst case happens when I try to install a new package in my OpenIndiana
   installation after a gap of say one month. Several minutes just to get a pkg
   of a few hundred KBs in size over my 8MBps link. Most of the time is spent
   in plan creation I think.

>
> On the other side, why take something completely new if there is experiences
> with a mature existing system?
>

   My point is the mature system is missing very key pieces like package
   management. Working on adding it will amount to re-doing a solved problem.
   My preference is to use a ready-made tested solution. In addition I simply
   do not like the SVR4 code. You can call it personal bias, but that is how it
   is.  I want SVR4 support but in a different way.
   I want to focus my efforts to better explore Pacman and see how to add
   Solaris platform and SVR4 support there. There are multiple options including
   an alien kind of converter:  http://joeyh.name/code/alien/
   So one can auto-convert from SVR4 format to Pacman format.

Regards,
Moinak.
-- 
================================
http://moinakg.wordpress.com/
_______________________________________________
belenix-discuss mailing list
http://mail.opensolaris.org/mailman/listinfo/belenix-discuss
http://groups.google.com/group/belenix-discuss

Reply via email to