...
| Having said that, IPS will put this right in the face of the end user, 
| so the poor packaging (too tight or too loose dependencies, packages too 
| fat/thin) prior to IPS cannot continue and still hit IPS goals.
| 
| > But later with IPS when I get to pick only packages I really want,
| > it'll be important to avoid bringing in large chunks of functionality
| > not everyone might want (e.g. installing apache shouldn't require
| > installing, say, postgres, because not everyone who wants apache
| > always wants postgres).
| >   
| 
| One thing I'd asked David Comay about with IPS, and he said had been 
| considered, is some sort of meta-cluster like functionality.  This is 
| important because the permutations with which these web stack components 
| and their extensions can be wired together is seemingly endless.
| 
| I don't think we can provide meta-clusters for all of the permutations, 
| but certain popular combinations (like A+M+PHP and A+PostgreSQL+PHP or 
| A+Ruby+MySQL) should be considered.  Asking the user to correctly find 
| all of the individual packages would probably be frustrating, but making 
| them strict dependencies and installing too much because we don't know 
| what run-time dependencies they'll actually use would be equally 
| frustrating.

If IPS provides clusters (in future), the problem of packages being too
thin may be a non issue. (use the cluster instead for the app that you want
to install.)

| One concern here, with pkg(5) being a no-scripting zone, something I do 
| agree with the design goals of, we may be left needing to consider a 
| post-install configurator or something of that sort.  Asking the user to 
| edit the httpd.conf/php.ini and know what to turn on in order to get, 
| say the postgres extensions turned on is something that would need to be 
| handled.

I would like to suggest having a central package per cluster for things
of that nature. (I might want to turn the php support on and off in
apache with out messing with the php support of - say lighthttpd)

so that rather than modify the httpd.conf on installation, it can wait
for the user to enable it. (and disable it if needed)

| I believe it would be impractical to make major changes to get the 
| upstream projects to work with SMF to describe how to wire things 
| together, though that would probably be the preferred approach with pkg(5).
| 
| I think I may have mentioned that Joe DiPol and I talked about what Java 
| ES is planning for, and that's their approach.  They have this 
| meta-cluster kind of issue too, as you may decide to install Portal with 
| Web Server or Application Server.
...
| > But for the above example, I wouldn't mind so much if things like
| > apache and apache modules are closely coupled at built time since
| > these are tightly related functionality anyway.
| >   
| 
| By the way, this is one of my frequently picked-nits with 
| blastwave.org.  I really like the work they've done over there, and I 
| support them (though Phil Brown and I may not agree on the pkg stuff), 
| but they too suffer from some of the package dependency issues.  For 
| instance, if you get Apache, it usually pulls down OpenLDAP, which then 
| pulls down all of it's dependencies (bdb), etc.  I suspect it's there 
| because of the build-time dependency, which is very different than a 
| run-time dependency.


                                    rahul
--
1. e4 _

Reply via email to