autopackage and other problems
Mike Hearn <mike@plan99.net>

• Was designer/maintainer of autopackage
• Am now Googler and not working on open source as much as before
• Generally controversial pest

• Distro centralisation doesn't scale
• Vendor lock in
• Currently use the AOL "walled garden" model. Neat, precise, small, sucky.
  • We want the web model - messy, sprawling, large, awesome.
• Ease of use (original reason)

• "Many ISVs want to use a single installer framework" - autopackage works on any distro
• "It is too difficult to install applications on Linux compared to Windows." (Jon Udell)
  autopackage designed to be as easy as possible: when it works, easiest/simplest experience of any platform
  but still much room for improvement in terms of denoising the ui
• "LSB needs to make it easier to do the right thing in this case (e.g., statitically link or bundle dynamic libraries).
  autopackage is "auto" for developers too - many tools to do binary analysis, relocatability, binary portability etc.
  Makes different kinds of linking trivial. These tools can be used separately from the packaging system.
• autopackage statistics framework let's you drill down and see why installs are failing (eg missing libraries)

• Revs often (every 6-12 months) to incorporate new technology, but is backwards compatible
• eg "Desktop Linux 1.0 compliant" apps should work on "Desktop Linux 1.5 compliant" distros
• Easy for devs to pick'n'mix features, ie app can run on 1.0 but runs better on 1.5 (really hard today)
• Large platform. Duplication, quality doesn't matter. It's OK to have 3 widget toolkits, 4 versions of the same API etc
• ONLY things that matter are:
  1. compatibility
  2. size

• Easy to create - YES (visual tools, automation)
• Easy to install - YES (one click inside a web browser, one click to confirm)
• Install location - YES (fdbr)
• License viewing at install time - KINDA (we have a tool to do it at first run)
• Single object - YES
• Easy dep resolution - YES
• Updates/upgrades - KINDA (no automatic online checking/download, but easy to add)
• Clean uninstall - KINDA (no recursive removal)
• Performance/memory footprint - NO (bash is slow), but not important IMHO
• Root access - YES (security is complex, ap can do either)
• Mass installation support - YES (headless mode)

• Distros don't care -- by far the biggest problem
• Long discussions with Ubuntu, Fedora, Debian, Gentoo etc devs leads to the same conclusion
• Every distro wants to be Number 1
• Few if any distros today are LSB compliant out of the box -- ISVs target what people use
• None are interested in trading off flexibility for compatibility
  • Guys at the top (eg Nat F, Mark S) tend to be more interested
• Often see themselves as being not Linux "there's a reason Ubuntu doesn't have Linux in the name"
  (so why bother trying to have unified packaging at all)

THERE CAN BE NO PROGRESS WITH THIS MENTALITY

• [insert usual complaints over lack of stability, etc]
• Linux is branding disaster-zone: user confusion, poor UI (skype website)
• ELF is broken and cannot be repaired (symbol scoping)

.... outlook not so good ....

• Complicated topic - talked for 20mins just on this at LUGradio Live
• Centralised distribution doesn't give security, just liability
• Thought experiment, "what if everybody did this?", where everybody == Microsoft
• Morally and legally questionable to appoint a bunch of random programmers judge, jury and executioner over ISVs
• Hard to prove or even define "evil" software in advance