| 1 |
Eric Jacoboni wrote: |
| 2 |
> Hi, |
| 3 |
> |
| 4 |
> - one of the main reason of my *BSD choice is about files locations |
| 5 |
> (when i ls a linux directory, it's like entering my son's bedroom |
| 6 |
> ;-). With portage system, i think it could be possible to choose the |
| 7 |
> target install directory, via an environnement variable (i've |
| 8 |
> searched but i've no find anything like that). With such a variable |
| 9 |
> it could be possible to assure that "system" ebuilds are installed |
| 10 |
> under /, but "non-systems" ones are installed in another place, say |
| 11 |
> /usr/local, if we want so. Seeing gkrellm installed under /usr/bin |
| 12 |
> is, definitively, not what i had choosen if i had to install it |
| 13 |
> manually. Furthermore, it could un-bloat /etc, /usr/bin directories. |
| 14 |
|
| 15 |
Why UNIX has the distinction between /,/usr ? |
| 16 |
I readed in the GNU Hurd documentation that it is just |
| 17 |
a tradition from the old tape times, althrough it is true that |
| 18 |
this separation eases the network sharing of binaries (/usr can be |
| 19 |
mounted readonly |
| 20 |
in remote clients). |
| 21 |
|
| 22 |
My system administration instinct says that I should install manually |
| 23 |
compiled software in /usr/local because it won't conflict with the packages |
| 24 |
managed by the system. This way, one can protect installed software |
| 25 |
and uninstall software manually without fear. |
| 26 |
But if some package management protects the config files, records the |
| 27 |
files |
| 28 |
owned by a package, handles concurrent version, and allows to uninstall |
| 29 |
at anytime, |
| 30 |
what does change if it is located in /usr or in /usr/local ? |
| 31 |
|
| 32 |
If you find it is important to keep the software phisically located in |
| 33 |
some partition rather than |
| 34 |
in some other I think that the best solution would be if portage would |
| 35 |
use "Stow" to store |
| 36 |
the whole package in /usr/local/stow/packagename and then symlink each |
| 37 |
element of its contains |
| 38 |
in / ... . This approach has the benefit of having concurrent versions |
| 39 |
and revisions of the same packages alive |
| 40 |
and hot-switchable. |
| 41 |
|
| 42 |
If you find more important to have a clean /usr/bin, I wonder if having |
| 43 |
a mess in /usr/local/bin would help. |
| 44 |
The "cleanest" situation is when you have 50% in /usr/bin and 50% in |
| 45 |
/usr/local/bin, since otherways the other |
| 46 |
will be a mess :-) |
| 47 |
The qpkg or epm utilities can help you in browsing your software |
| 48 |
better than a "ls /usr/bin" with 3000 lines output will do. |
| 49 |
|
| 50 |
I agree that it would be nice if the administrator could choose wich |
| 51 |
philosofy to follow, but |
| 52 |
consider that many ebuild will have to be rewritten and/or perhaps |
| 53 |
filled with hacks to implement |
| 54 |
this relocability, because some software installation is not the simple |
| 55 |
/usr/bin -> /usr/local/bin thing. |
| 56 |
Consider for example "non system" software which installs kernel modules |
| 57 |
(for example AVFS, a user mode |
| 58 |
filesystem, or LiS, linux streams implementation. both are not yet |
| 59 |
complete software and cannot be threaded |
| 60 |
like "system" of "base" installation). |
| 61 |
|
| 62 |
Just my 2 cents, |
| 63 |
|
| 64 |
Marko Mikulicic |
Archives