| |||||||||||||
|
| |||||||||||||
New approaches to Linux package management (Linux.com)New approaches to Linux package management (Linux.com)Posted Nov 11, 2005 18:35 UTC (Fri) by mmarq (guest, #2332)Parent article: New approaches to Linux package management (Linux.com)
I look to Conary with a whole lot of hope.
Its gonna be this time that open-source is gonna mean something to the average user!?. Is it possible that Conary evolves to a "graphical" method where a user installs from source a package with only a few clicks!?. The actual RPM and DEB being them from source or not dosent make a real difference to the final user.
The possibility of having an assisted "compile and install the source" of a whole distro from a safe repository, or Gentoo on stereoids, can prove to be a terrible marketing tool.
With broadband going to standardize arround 4 to 8 Mbs, distros should being much better offering assistence to compile from source to everybody, than trying differentiation upon features rushes, many times with non standard stuff, only to beat the competition!
Who happens to get the best service will win, and not who happens to get a particular feature only usefull for a little tiny percentage of users.
(Log in to post comments)
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 1:29 UTC (Sun) by drag (subscriber, #31333) [Link] Well I don't know much about the conary approach or anything like that...
But it's a pretty straight forward thing to simply recompile debian packaged software for whatever optimizations you want. That's what all those deb-src things are about, after all.
The reason people don't do this more often is because it's franky nearly completely pointless to recompile packages for optimizations and can actually slow things down.
For instance kernel developers discovered that using GCC's optimizations for kernel slowed things down for the kernel.. not because it's bad optimizations, but because the resulting binary size is large. If they optimized for size then it resulted in smaller binaries and led to less memory fragmentation.
Only a very small amount of software can have any benifit anyways... and then it's not that big.
What somebody needs to do is figure out a magical way to take any source code you want, dump it into a universal magic-compile magical app that will compile it, install it, and then allow you to uninstall it, plus magically handle dependancies.
Anything short of that your not going to see any improvement over the current Deb model.
The only way to get package managers better is to sit down, and get people to actually figure out a common package format and make this package format common to all distros, and figure out a universal way to handle dependancies. All it takes is a lot of hard work and time on the part of package developers and getting people to agree.
Which I doubt is going to happen any time soon.
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 3:20 UTC (Sun) by bangert (subscriber, #28342) [Link] to me optimization is not about compiler flags...and as you say yourself, binary sizes are important for speed as well. i want to get rid of all the (compile- and run time) dependencies that i don't use AFAIK debian can't give me that. i had to work on a SUSE system the other day - gosh, the dependency tree for the base system is incredible. half of the stuff, i would never need... granted - source based distributions have a pretty large base system dependency too, needing a compiler and all...
New approaches to Linux package management (Linux.com) Posted Nov 14, 2005 9:40 UTC (Mon) by ranger (guest, #6415) [Link] and as you say yourself, binary sizes are important for speed as well. i want to get rid of all the (compile- and run time) dependencies that i don't useAFAIK debian can't give me that. But, at least when you find out you *do* need the feature, it doesn't take you almost a day to rebuild all the software with support with the specific feature.
New approaches to Linux package management (Linux.com) Posted Nov 17, 2005 17:57 UTC (Thu) by hazelsct (guest, #3659) [Link] Yes it can:
apt-get build-dep mozilla
First "apt-get build-dep" downloads the mozilla source package and all build dependencies (library -devs, compilers, etc.) and builds everything.
Granted, it's not point-and-click, but again, why would anyone sane build from source anyway? (Except kernel images, where subarch customizations and size can be important.) There was a great benchmark a couple of years ago which showed Gentoo running many tasks several times slower than Debian and RedHat because of misguided excessive optimization.
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 5:35 UTC (Sun) by bk (guest, #25617) [Link] What somebody needs to do is figure out a magical way to take any source code you want, dump it into a universal magic-compile magical app that will compile it, install it, and then allow you to uninstall it, plus magically handle dependancies.It's called Portage.
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 10:37 UTC (Sun) by drag (subscriber, #31333) [Link] Realy?
What Portage seems to me is more like RPM spec files, just without the RPM part.
It still takes somebody to put the effort into testing and producing the package and making sure that dependances are taken care of. If a package is not in portage or it's not up to date it's just as useless as anything else. There is nothing magic about it.
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 11:19 UTC (Sun) by bk (guest, #25617) [Link] Any standard autotools-enabled source tarball can be installed via Portage with a standard ebuild template. You do have to tell it about dependencies, however. Think about the alternative, though: Do you want your application's dozen different dependencies all included in the same package? What happens if you want to uninstall the package but other programs depend upon that package's dependencies?
In practice the vast majority of a package dependencies already exist within the Gentoo Portage repository, so it mostly is just a matter of plugging the package into the ebuild template and installing (and submitting your ebuild to Gentoo's bugzilla so others can benefit from it).
New approaches to Linux package management (Linux.com) Posted Nov 14, 2005 0:57 UTC (Mon) by drag (subscriber, #31333) [Link] Well I found out from using Gentoo that the majority of the time I compiled to loose dependancies, a few weeks later I'd want them anyways to enable different features. The net result was me wasting my time recompiling the same software stack twice...
When it's something like Gnome or involves something like OpenOffice.org your looking at several hours worth of crap you'd have to wait on.
So with the extra dependancy issues.. I don't care anymore. At most your saving a few dozen megs of harddrive space, if you don't use it or need it then it just takes up harddrive space and has little effect otherwise. In practice the benefits are negligable, IMO.
If you uninstall a package with Apt and other programs depend on it's dependancies then the dependancies aren't removed to suit the other programs. It's pretty simple. If you do end up removing packages and have stale dependancies then there are things like deborphan and whatnot to check for that. Worst case you just have to go thru aptitude or synaptic and pick and select items you don't want around anymore.
The portage system is nice.. but as far as things go for quality control, package compatability, having all the software a end-user wants, keeping things up to date, and other issues that are universal to all package management it still requires as much brute man-hours as anything else.
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 15:57 UTC (Sun) by mmarq (guest, #2332) [Link] " What somebody needs to do is figure out a magical way to take any source code you want, dump it into a universal magic-compile magical app that will compile it, install it, and then allow you to uninstall it, plus magically handle dependancies. "
Perhaps i'm dumb presuming that a lot of things are sub-understood, but that is exactly what i had in mind. The exact form and details ?... as everything in computer science, i'll bet that there always be more than one possible way.
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 1:43 UTC (Sun) by man_ls (subscriber, #15091) [Link] Is it possible that Conary evolves to a "graphical" method where a user installs from source a package with only a few clicks!? The actual RPM and DEB being them from source or not dosent make a real difference to the final user.There are currently a lot of graphical installers, why is Conary so important in this respect? Its strengths lie elsewhere. By the way, the very existence of "end users" is not undisputed; many people think they are "uneducated contributors". The term is thus a bit derogative. The possibility of having an assisted "compile and install the source" of a whole distro from a safe repository, or Gentoo on stereoids, can prove to be a terrible marketing tool.Gentoo "on steroids"? What do you mean? Right now you can "compile and install the source" (I assume you mean "install from source code") with Gentoo. It is a "terrible" (as in "terribly bad") marketing tool: the unwashed masses don't care at all. With broadband going to standardize arround 4 to 8 Mbs, distros should being much better offering assistence to compile from source to everybody, than trying differentiation upon features rushes, many times with non standard stuff, only to beat the competition!What a terrible mess of a thought. Why is bandwidth so essential to compile from source? You need to download it, true, but then you need the tools and the processor time to do it. Why is it better than downloading binaries as with Debian or Ubuntu (which are not that much smaller in download time anyway)? Who happens to get the best service will win, and not who happens to get a particular feature only usefull for a little tiny percentage of users.I don't see what service has to do with bandwidth, compiling from source, graphical installers or distro features. Or with Conary. It is too hard to follow, sorry.
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 16:48 UTC (Sun) by mmarq (guest, #2332) [Link] Geez man_ls... calm down a bit will you ? i appreciate your comments,i honestly do, though they seem to be 'a priori' totally contrary to my opinions.
Following your comment, i'll try to be very short.
a)Its not derrogative... why a prejurative sense in here? and to whom?. The general masses of users simply dont care about anything 'non graphical' and conary will simply be perpetually unfinnished without it, and finnished if it adopts its own, synaptic, smart or autopackage or...
b)Conary is simply better, in every sense, than portage. The unwashed masses are , *simply_too_simply* the base of any long term, sustainable and dominante IT paradigma. Microsoft has taught that too many times. And it can only be a *terrible good marketing* if someone offers the possibility of having optimized configurations of systems with a very strong tone on the personalized item(like in cars). Advanced users simply dont buy distros, they scratch their own itches, and personalized is something that i would definitely buy.
c)Apart from what you have answered to yourself, the rest is the same as above.
d)Simply it has to do with Linux as a service a not as a product. Everyone seems to want their own distro, and if distros offer assistence for that possibility out of their safe and tested repositorys(as in a service), instead of trying to sell an assemblage of packages, they will be much better.
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 17:46 UTC (Sun) by man_ls (subscriber, #15091) [Link] Geez man_ls... calm down a bit will you ? i appreciate your comments,i honestly do, though they seem to be 'a priori' totally contrary to my opinions.I would rather say "orthogonal", up to the point that I can understand. The things that get you excited leave me cold as a stone; you get starry eyed while I yawn. I have nothing against you. Sorry if I seem to have little patience with your comments, but I would like you to articulate your thoughts a little better in the interest of mutual understanding. Its not derrogative... why a prejurative sense in here? and to whom?"End user" is derogative in the sense that it makes the user a passive consumer, an idiotic entity just takes whatever is given to him or her and never gives anything back. This is precisely what Microsoft and their ilk do; and it is precisely what they do wrong. Maybe it's the only way to make money out of computers; but if it is, then I don't want anything to do with it. If free software is to make a difference, then it will be to teach these masses of users that they can do something about it. And we have seen this approach work in a number of ways. Look at Groklaw; it has taken a mass of intelligent but technically naïve people and has made them into a loyal audience that will collectively analyze every bit of information that goes throught them. if distros offer assistence for that possibility out of their safe and tested repositorys(as in a service), instead of trying to sell an assemblage of packages, they will be much better.Still, nothing to do with source code. Mandriva offers such a service (it was "Mandrake Club" before, "Mandriva Club" now I guess) with the possibility to download and compile your own packages. What does Conary have that is better? Less downloads? You pointed out in your previous message that bandwidth is cheap now, so?
New approaches to Linux package management (Linux.com) Posted Nov 13, 2005 22:15 UTC (Sun) by mmarq (guest, #2332) [Link] " Still, nothing to do with source code. Mandriva offers such a service (it was "Mandrake Club" before, "Mandriva Club" now I guess) with the possibility to download and compile your own packages. "
Everything to do with source code. It was always possible, before Mandrake Club, to download the code(SRPMs) and build the packages, i've done plenty of those.
But it isn't, even a little bit, easy for general end users. Conary promisses making that and specialy upgrades from source code more easy.
source code Posted Nov 14, 2005 11:50 UTC (Mon) by man_ls (subscriber, #15091) [Link] Conary promisses making that and specialy upgrades from source code more easy.I will make the questions more obvious. Why is upgrading from source code so important? What does it give you that you cannot get from Debian's APT, or Red Hat's RPM? If source code is really important, why is Conary better than Portage? Simply because it says it is "advanced", or are there any features that appeal to you? Otherwise, you might as well ask for a graphical interface for Portage.
source code Posted Nov 15, 2005 0:53 UTC (Tue) by njhurst (guest, #6022) [Link] Source might get you slightly closer to the bleeding edge (although most big projects these days carry a debian dir around with them anyway). I agree though, in practice source is really just a waste of time. And that's coming from someone who uses Gentoo and Debian.
Regarding graphical interfaces, any idea what's wrong with say synaptic?
|
|
||||||||||||