I mostly disagree with the OP’s analysis.
Problem 1: ‘Difficulty distinguishing OS from apps due to package management’
This is an education problem, rather than an OS problem. Many people using OS X & Windows (and other OSes that employ universal installers) don’t know where the OS/App divide is either. They don’t know Garageband is an app on OS X rather than a part of OS X nor can they tell you where Windows ends and Microsoft Office begins. I see salespeople move units of MSO all the time because consumers are shocked, “you mean it doesn’t come with the machine?!?” Just like how many “open the internet” rather than recognising they are really opening an internet browser.
Many people using a computer see a unified appliance that happens to do different things (like a swiss army knife), they do not see the reality that it’s really a hardware + software stack with each layer providing difference functionality to create the whole experience. In just the same way they see a toaster (even a washing machine!) as a unified appliance, not a self-contained electronic/mechanical system with interdependent sub-systems which combine to provide the intended function(s). People recognise that there are ‘parts’ to mechanical devices if they stop to think about it, but in everyday living it often does not occur to them. I’d bet more people would recognise the fullness of a washing machine’s ‘part-ness’ than that of a computer system when pressed to think about it - “the window boarder thingy is a separate program? Really?”
Problem 2: ‘3rd Party support is poor’
Many 3rd parties are looking to make money. It’s true that market fragmentation hurts adoption because there are not strong enough standards, but the companies that think they can make money on Linux provide the software. Take Wolfram’s Mathematica, it’s pretty serious software - and it’s available on Linux. Why? There’s a large enough market of academic and business users on Linux to justify it. They even publish a version for Solaris.
Any other vendor could do the same and if desktop Linux marketshare rises in their niche, they will. Right now the marketshare is seemingly so low (especially in countries with stronger cultural norms of paying for software) that many 3rd parties don’t bother because it doesn’t pass cost/benefit analysis. I don’t doubt a unified installer would help, but I do doubt dropping a unified installer on such small marketshare would start a frenzy of porting apps to Linux.
Problem 3: “Must track dependencies on uninstall”
I can understand this frustration, but I think the concern is a bit overplayed. In the case of universal installers, apps have to bring their own dependencies with them. This means if you installed Pidgin and GIMP you’d also have GTK installed twice, add AbiWord and you get it a third time and on and on. So in the end you’re likely to have more wasted space under a universal installer system than the shared library system Linux typically employs.
A default openSUSE 11.1 install is around 3GB, a default Windows 7 install is 6GB+ and by default lacks much of the functionality openSUSE has out of the box. This means you’d need to accumulate at least 3GB of dependency cruft before reaching the size of a bare Windows 7 install (which itself is smaller than a bare Vista install).
In any event this is a bit of non-issue since deb-based systems have the deborphan program and rpmorphan is in development. Additionally, some package management teams are working on dependency clean up as well.
The Solution: ‘Employ a universal installer to achieve the same level of usability’.
Let’s not mistake “familiarity” with “usability”, we could go round and round about which is better. Personally I think I could persuade many users that a “Google for getting applications” that lets you update everything from one place is a better idea rather than running from site to site getting apps piecemeal, having a number of updaters in the background slowing the computer down, and getting different looking pop-ups from each updater asking for permission seemingly at random.
It’s all in how you present it. And besides, outside of Microsoft Office there aren’t that many unavailable-on-Linux apps that the typical user actually uses. “But what about Photoshop?” you say? In 2008, the average desktop PC cost $550 and the average laptop cost $700 (USD). Adobe Photoshop CS4 costs $700, even the student discount price is $300 (USD).
Sure, the inability to run a pirated version of Photoshop CS4 probably hurts Linux to some degree. I suppose the software thieves will just have to settle for CS2.
In any event, universal installers do exist in the FOSS universe. The klik system has been around for a couple of years, PC-BSD employs a universal installer method as well. Neither of those projects wound up capturing huge adoption, nor did 3rd parties rush to ship their apps via those methods.
So after all that…I’m not saying a universal installer would be totally useless, indeed it has some strengths, but I don’t think those strengths outweigh the weaknesses. If we look at the cellular market, the wild-west-rope-an-app-from-different-fields method has been completely displaced as the standard by the ease of centralised “app stores”. I doubt this ‘new’ distribution method is going remain ‘only on smartphones’ forever. And while it could be argued the apps carried do not have dependencies in the same sense as Linux apps, at the very least we can agree it’s a mark against the idea that ‘centralisation blurs the line and that’s bad’.
Ok, I’m done now.