The Future of Packaging Software in Linux

michuk writes "There are currently at least five popular ways of installing software in GNU/Linux. None of them are widely accepted throughout the popular distributions. This situation is not a problem for experienced users — they can make decisions for themselves. However, for a newcomer in the GNU/Linux world, installing new software is always pretty confusing. The article tries to sum up some of the recent efforts to fix this problem and examine the possible future of packaging software in GNU/Linux."

Comments

Applications Packages

[SultanCemil]

If only major linux distros would use Application Packages like OS X, the world would be a better place.

Seriously, drag-n-drop installation rocks.

Re:Applications Packages

[croddy]

The reason Linux distributions have not been trembling to adopt the OS X style of package management, if you can call it that, is that it would be a poor fit for the Linux software ecosystem.

The vast majority of software used on Linux systems is licensed under the GPL; what is not is almost always under another license permitting free redistribution. This gives Linux distributors great freedom in selecting and assembling a compatible collection of versions, tested and working with the same versions of dependent libraries. In a larger distribution (such as Gentoo, Debian, or Fedora), most of the software you will ever need is already a part of the OS -- you just need to use the built-in package management tools to summon it from the distributor's repository.

OS X-style package management is best suited for a software ecosystem in which users draw software from a large number of heterogenous third-party sources, while the core OS and iLife suite are maintained and updated by Apple. A third-party distributor who wishes to distribute something that must link against a particular version of a library can include it in the application bundle, knowing that the exact version needed will be available. This can lead to many copies of the same libraries being installed, facilitating compatibility with applications that require different versions, but consuming (small amounts of) disk space unnecessarily and increasing the attack surface when multiple copies of an exploitable library are installed on the system. A system such as APT does not need to provide a facility for private copies of libraries, since it does all of the dependency computation, and all software in the repository is built and linked against the libraries in the repository.

Certainly, once you have resigned yourself to visiting a third-party distributor's web page, manually downloading a binary package, and then manually installing the binary package, drag-and-drop installation is very convenient. But the Linux software ecosystem does not require this concession from the user -- the Linux distributor is free to provide a repository and tools for finding, installing, and updating software, without the need for manual installation.

Re:Applications Packages

[AKAImBatman]

As does clicking on a checkbox. Seriously, I don't need it to be any easier.

Unless the software you want isn't in the Synaptic repository. Then it's hell on earth for the average user. The only response they get from support and developers is, "Why would you want to use software that isn't in the repository?"

Actually, that's not true. There are plenty of other fun responses:

"You should compile it from source."
"The vendor should spend his time getting his software added to our respository!"
"Use RPMFind. I'm a developer and I've never had a problem installing binary packages on the distro I work on." (Conveniently ignoring that when something breaks, the "developer" fixes it himself.)

Not that there's much point in harping on this again. I'll just get the same, "U R STUPID", "You need to try distro XYZ", and "Everything is in my distro's repository!" answers I've gotten before.

Blinders on, and full speed ahead cap'n!

Hasn't explored other packaging methods

[shermozle]

(while discussing RPM)

Still, a lot of other systems like Ubuntu, Debian, Slackware, Gentoo or Linspire do not use the RPM format and do not plan to incorporate it. What that means is he hasn't used any other packaging formats. Common mistake that people think RPM is somehow "best" because it's used by a few distros. Do some searches for "circular dependency RPM" to see why that's just not true.

Re:Hasn't explored other packaging methods

[DoubleRing]

Hear hear! I have mod points, but I'd rather post.

Circular dependencies, aka RPM hell, is what actually prompted me to make the switch from the Red Hat family to the Debian family. I used to be a pretty die hard Red Hat user. It used to be that Fedora was the cutting edge, back in the core 2 and 3 days. I would have those days when I would wrestle with the packages, but I just took my hits and moved on. Then Ubuntu came along, and I realized how much time I was wasting with that stuff. It "just works." APT is great (it's a pity POSIX decided to go for RPM). Gentoo's portage is really cool too, but IALAB (I'm a lazy bum--if you can't reconcile the acronym, then you probably shouldn't know what the missing word is).

Re:The solution!

[Max Littlemore]

Using multiple package formats is great idea, IMO. I use alien on Ubuntu for those situations where the software I want is only avaliable in RPM, but as it says in the summary, new users can be a bit confused by this and building from sources is often too much. I would like to see GUI tools get the smarts to automatically figure out dependencies across all formats, allowing all distros to become package agnostic. Perhaps Linspire's CNR interace would be a good candidate for this.

Also, the option to resolve dependencies and install as a statically linked blob would be awesome for legacy stuff. I've lost count of the number of times I've wanted to install an app, only to find that it relies on some obscure version of xyz.so and won't work, so I find the source for the old version of xyz, only to find it depends on some older version of abc.so. If I could get this xyz.so, etc without conflicting with that xyz.so, create a static binary and put it somewhere under /opt, I'd be happy. I know it's not elegant, and that it uses more storage, but as a work around for difficult to support stuff, it ain't so bad when storage is cheap. Some apps I always install as blobs anyway, such as blender.

BTW, from TFA: Network Access Message: The page cannot be displayed
Slashdotted :-(

A total load of bullshit, and here's why

[Overly Critical Guy]

My favorite thing in package discussions is when someone with an agenda towards a particular implementation writes out a list of steps. The alternative is always padded with extra steps to make it more difficult-looking while their favored implementation is reduced to look squeaky clean and easy.

You padded the Mac list with the following:

• "Open disk image that contains the program." - DMGs are auto-mounted by Safari.
• "Open Applications folder." - There's already an Applications shortcut on the Finder, so you just drag to that when the disk image window automatically opens.
• "Create new icon in dock." - The fuck? You don't have to do this
• "Have to recheck the site periodically to check for a update for a specific program" - Bullshit. This doesn't even have to do with package management, and it's an OS X convention for apps to auto-check for updates when they're run. You don't have to recheck any websites.

Your Debian list conveniently leaves out having to click the KDE start menu, fire up a Terminal window, type in the root password, waiting while the package manager goes through dependencies, etc. What a phony comparison of steps. I could just have easily reduced OS X's step to one line of "Drag app icon to Applications shortcut" in the same the way you reduced Debian's steps.

Re:Nonissue

[zzatz]

"...never make a significant dent in the Microsoft+Apple consumer desktop market."

Linux will never make a significant dent in the Microsoft+Apple market by doing the same things the same way as Microsoft and Apple.

Look at markets where Linux has succeeded, such as servers and embedded systems. Linux succeeds *because* it doesn't follow the Microsoft license model, the Microsoft development model, the Microsoft business model, and so on. You can't win if you play by Microsoft rules.

Linux can be, and is, an OS for users. It isn't an OS for third party closed source binary distribution. Don't read that as non-commercial; commercial software was distributed in source form before Microsoft and will be again. Distribution in binary form makes sense for games and art, but not for general purpose computing. The value of doing things in software rather than in hardware is that software is malleable. But you need the source to realize the full value; binary distribution removes value.

So yes, Linux will not make a significant dent in the Microsoft+Apple consumer desktop market, if that means the closed binary sales market. If Microsoft played in the NFL, they'd be the Super Bowl winning Colts. But the Colts will never win the World Cup, which is worth more. Don't complain about Linux not hiring a bigger front line when the game Linux is playing is soccer, and doing rather well at it.

Re:The solution!

[nmb3000]

The real bastard is that each distro has subtle differences in how the packages and the dependencies are organized. The only way that I can see to fix that is to design a universal package tree, and convince all the major distros to conform to it. Which is not impossible, but it aint easy, either. And it might cause other problems.

Which is why, as it currently stands, this year will not be Year Of The Linux Desktop. Consumers won't just accept that they can't install software X because it's an RPM and alien doesn't work (this is of course after looking online for half an hour to figure out that alien is the tool to use). Manually compiling from source is simply not an option for standard users. Sure it's a dandy idea, and if you get a "fullproof" GUI that handles the compilation and installation then maybe, but I can't count the number of times make/make install has failed for some obscure reason. The first time grandma needs to go download dependencies means Linux has failed on the consumer desktop.

This is one place that Microsoft and Apple have it right. By having a standardized method of installing and storing program information they make getting new software many times easier than on Linux (excluding the "normal" packages. I'm thinking more along the lines of tools and apps you download from the web). This is also one reason people are willing to pay for an operating system that has a standardized and dependable way of doing things.

Microsoft even released the WiX toolkit that allows anyone to create MSI installer packages. MSIs are one of the best ideas for Windows in a while: No more dealing with poorly-written homebrew installers or 10-year old, 16-bit InstallShield programs. Instead you have a fully scriptable installer that's transaction-based and has near 100% support coverage.

I like apt, but downloading a gzipped file of source or a deb that complains about dependencies still can't compare to an MSI package. Even if a solution was developed that worked as well as or better than MSI, as you say, it would take significant effort (and maybe not even then) to get it supported by all the major distributions. Some people seem to think that the fact that Debian does things differently from Mandriva that does it different than Fedora is what makes the distribution "special". Be that as it may, I think it's only hurting Linux users as a whole.