Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 24 Apr 2002 14:15:05 -0400
From:      Antoine Beaupre <anarcat@anarcat.ath.cx>
To:        Mike Meyer <mwm-dated-1020103973.b26615@mired.org>
Cc:        freebsd-libh@FreeBSD.org
Subject:   Re: packaging base
Message-ID:  <342BD734-57AF-11D6-AE88-0050E4A0BB3F@anarcat.ath.cx>
In-Reply-To: <15558.62884.852620.270991@guru.mired.org>

next in thread | previous in thread | raw e-mail | index | archive | help

Le Mercredi 24 avril 2002, =E0 02:12 , Mike Meyer a =E9crit :

> In <0DFF2010-57A8-11D6-AE88-0050E4A0BB3F@anarcat.ath.cx>, Antoine=20
> Beaupre <anarcat@anarcat.ath.cx> typed:
>>>> And libh will meet resistance not only from being a brand new =
system,
>>>> but also at trying to package base, which will break havoc among
>>>> developpers.
>>>
>>> How many developers use sysinstall, vs. rebuilding from source? =
Those
>>> are the only ones who are liable to care. If it's done right, then =
the
>>> new sysinstall should have packages defined by the NO* variables in
>>> /etc/defaults/make.conf, and should set the appropriate flags in
>>> /etc/make.conf for each part you don't load.
>> Please no. Please let's get rid of those variables. Please lets just
>> seperate the different parts of the tree clearly and isolate their
>> dependencies and let the developper make install where he wants. =
Using
>> variables, we'll end up with hundreds of them. It will be a =
maintenance
>> nightmare.
>
> Now you're talking about breaking "make buildworld", and that will
> generate a lot of resistance. It's not clear what you're proposing
> replacing it with, except for some portupgrade-like utility.

Yeah.. I guess I just threw this as a thought. Forget it. :)

>> installworld is somehow doomed to go in the new scheme, as everything
>> will be a package and the line between base and ports will be =
blurred.
>> Everything installed through this procedure will have to be =
registered
>> through the package system.
>
> Yes, everything needs to be registered. No, installworld doesn't have
> to go away. I can see an installworld target that "knows" what
> packages are part of the base system, and only installs the ones that
> are already installed. That's actually cleaner than using make.conf
> variables. Buildworld can use similar tactics.

That's a very interesting idea.

That's why developping pkgAPI is so important: there will be a=20
transparent way of getting this kind of information.

> But that's all *very* vague. A solid proposal is in order. Since
> you've apparently done more thinking on this than me, do you have one
> in mind?

*Nothing*. I've been thinking about this long and large, and I didn't=20
see any clean way of packaging the base system apart using the ports=20
collection which obviously, isn't really a good solution.

> This is potentially something I can work on.

That's why I'm bugging you with this. ;) I just want you to avoid=20
thinking that libh is some kind of silver bullet that would take care of=20=

this. It's not taking care of this.

packaging of the base system, once done, will easily be integrated to=20
libh. I suggest you start attacking the problem by making abstraction of=20=

the floppy problem. I see it as just a technical quirk.

I'd like to see a finer-grained packaging than just one big "bin"=20
though. :)

> Libh isn't, as I what little I know of tcl is enough to keep me from=20=

> wanting to learn more.

libh is more than just tcl. There's a solid C++ API behind all this.

> However, something like tcl is required, because part of the new
> port/package system is a safe way to encode actions on packages.

yeah. TCL or another sandbox thing to allow safe execution of packages.

a.


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-libh" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?342BD734-57AF-11D6-AE88-0050E4A0BB3F>