gnu-system-discuss
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: GNU System Explanation


From: Barry deFreese
Subject: Re: GNU System Explanation
Date: Sat, 11 Feb 2006 22:19:56 -0500

----- Original Message ----- From: "Filip Brcic" <address@hidden>
To: <address@hidden>
Cc: "Barry deFreese" <address@hidden>
Sent: Saturday, February 11, 2006 9:40 PM
Subject: Re: GNU System Explanation

Дана Saturday 11 February 2006 19:56, Barry deFreese је написао(ла):
Can someone please give me a concise explanation of what the intended goals
are here?

The goal is to make package installation and removal as easy as cp and rm.

This is my point. I don't think it can be unless you ignore things like dependencies.

I understand that you want packages isolation in /packages, but why?
Managing multiple versions of a package could potentially become a headache
I believe.

With current state yes. But that problem will be solved.

How?

For example:  I have /packages/foo-1.1/bin/foo  and /lib/libfoo.  Then I
install foo-1.2 which has /packages/foo-1.2/bin/foo and /lib/libfoo,
foo-1.1 becomes pretty much unusable anyway, does it not?

Why do you need both foo-1.1 and foo-1.2?

I would ask the same question. If this is not the goal, what is the advantage of "package isolation"?

If you take a look at how Mac OS X implements several versions, you can see
that you have

/path/to/foo/1.1
/path/to/foo/1.2

and

/path/to/foo/Current (symlink to for example 1.2).

That is not a bad way to handle multiple versions. It would, of course,
require tweaking of stowfs, but... that should not be the problem. And that
way you can use 1.1/lib/libfoo and 1.2/lib/libfoo if you explicitly need
libfoo by version. If you don't care about the version, use /lib/libfoo and
you'll get the current version.

OK, understood but this still seems extraneous to me, the uninformed :-)

I also understand that you have philosophical differences with Debian.
Fine, but Debian provides a nice set of tools for package management when
dealing with things like dependencies, reverse dependencies, etc. Yes, it
would be nice to be able to use common filesystem commands to install and
manage packages but is it realistic? If I extract foo-1.1.tar.gz to
/packages/foo but foo needs bar to run now what?  OK, so I extract
bar-1.0.tar.gz to /packages/bar. But now I install baz which also depends on bar, there becomes 1 to many and potentially many to 1 relationships of
packages.

I didn't quite understand your example. Dependencies will be implemented
somehow.

This is my fundamental question. If there are decent workable package management systems out there, why not use them rather than trying to re-invent the wheel?

What do people see as currently wrong with dpkg other than being Debian and
a little cumbersome for developers?

Why cumbersome? I find it much more developer-friendly than for example rpm.

I like dpkg/apt, I just hear from lots of "GNU" folks that it sucks. Heck, right now if stowfs is fast enough I can take a Debian package and do apt-get Dpkg::Options::="--instdir=/packages/foo/1.2/" . Right now it breaks because the postinst scripts and such try to run in "standard" directories so it breaks. However, if stowfs is "immediate", this might alleviate this problem.

Not to mention, there is always the option to modify dpkg and all the debhelper scripts to do what you want.

Thanks for the response!

Filip


Barry deFreese (aka bddebian)




reply via email to

[Prev in Thread] Current Thread [Next in Thread]