On Monday 03 October 2005 12:47 am, Ciaran McCreesh wrote:
> On Sun, 02 Oct 2005 21:50:07 -0400 Dave Nebinger <dnebinger@joat.com>
>
> wrote:
> | Their build system suits their purpose - distribute a LAMP-like
> | system for the foundation of their web application. I'm sure it will
> | keep them from getting distracted from questions like 'zimbra works
> | for postfix 2.2 but breaks for 2.2.3'; they provide it all and to use
> | it you're normally stuck with their 3rd party binaries at the
> | version/patch level they give you.
>
> Hrm. Does this really need an ebuild?
I think so. As a stand-alone package it has a number of dependencies that
need to be managed, something Portage handles quite well.
Also, I'm finding that at it's core, portage handles the reorganization of a
distribution quite nicely. Being able to automagically handle the java
dependencies using java-pkg is a godsend. Itematically choosing the
individual files from the distribution build and retargeting them
to /usr/share/webapp with a one-line command... The advantages appear to
outweigh the costs.
> Wouldn't it be better to use the
> associated portage-provided packages?
That's the goal - strip out the parts from their distribution that have
exiting components already within the portage tree (and already installed on
my system). Boils down to a couple of wars for the most part, plus scripts,
things like that.
> Also, how do you intend to handle
> security updates?
For the most part I'd leave that to Portage for the 3rd party components. I'm
expecting that once I get the initial ebuild figured out, when they release a
new distribution I'll just have to re-diff & re-patch if the current patches
stop working.
> | > | a /var/db/pkg query system,
> | >
> | > Yick! Bad bad bad idea.
> |
> | Yeah, I know. But how else do you answer the question "Hey, Portage,
> | where did you really install that my.cnf file?"
> |
> | Obviously the system admin is free to move the my.cnf file or even
> | use a different file/path altogether. But at least it would give me
> | a starting value to use at compile time...
>
> I'm starting to think you really shouldn't be ebuilding this lot at
> all...
I understand your doubts, and trust me I have them to. But the way I see it I
have two basic choices:
1. use their distribution. This discards all of the advantages that portage
provides (i.e. updates to the foundation software). Also means that for each
component in their system that is already on mine, I need to shut mine
down/unmerge them so the system will rely upon theirs. Will really mess up
things like openldap and mysql dependent projects as their distribution
doesn't provide the full development stuff portage would need to handle
builds.
2. create an ebuild to merge only the necessary components into my system,
taking advantage of the fact that I already have working components
installed.
It would seem to me that #2 is the appropriate gentoo-way...
I started working on this because I was initially looking for a good web mail
package to use on my gentoo box (I'm starting a new contract where I won't
have pop3/imap access to my email from the job site). Scope soon creeped to
become a good collaboration suite to use on my gentoo box (because I'd have
calendars, address books, etc. scattered across many different sources).
While I was checking out the options that were already in portage I was also
checking online for other possiblilities. The recent posting to slashdot on
the very subject caught my eye, and when I went through the flash demo at
http://www.zimbra.com I was really impressed with everything I saw.
But to get it up and running I soon discovered I'd have to make one of the
choices from above. Eventually some 1,000 year old geezer might say "He
chose poorly," but who knows - it might work and might catch on.
--
gentoo-dev@gentoo.org mailing list