Re: AutoPortage

Hi guys.

I have to agree with Matthew on this one. Autopackage scales in theory, 
but in reality it has been less than successful. I think we're too 
caught up with trying to fight the problem with technology. It's 
fundamentally more of a social problem, not a technical problem, in that 
it *can* be solved with technology, but isn't, because everyone's 
unwilling to change their mindsets.

"Packaging as separate from development tends to introduce obscure bugs"

Yes, there are a few real-world examples of this. But the developer 
won't necessarily do a better job at packaging, especially for distros 
that he's not familiar with. I think the chance that the developer 
screws up is less than the chance that an independent packager screws 
up, but that all becomes irrelevant if the developer is unwilling to 
create packages.

As much as I regret saying this, I don't use Autopackage for own 
software (www.modrails.com and www.rubyenterpriseedition.com) despite 
having been an Autopackage developer. Instead, I wrote custom 
cross-platform installation scripts which  1) helps the user with 
compiling the sources and  2) checks whether all necessary build 
dependencies are installed. I leave packaging to third parties because 
it's too much of a hassle to deal with by myself. At the same time I 
actively accept packaging patches in order to ensure that downstream 
doesn't break things with their own patches.

The mindset of the user base has changed very little over the past few 
years: the hardcore users still prefer their distro's native packaging 
system. In fact, I'd argue that more and more users have become 
"addicted" to their native packaging system, not excluding myself. 
Software installation on Ubuntu works remarkably well in 99% of the 
cases. The 1% is still painful but not painful enough to throw apt out 
of the door. Especially Debian/Ubuntu users are fanatical about having 
everything installed through apt.

Interestingly enough, I've found that the combination of an installer 
that compiles from source, combined with just an Ubuntu 8.04 package, 
works remarkably well and covers 99.5% of the installation needs of my 
user base. Nobody complains about installation of my software - in fact 
everybody praises us how easy it is. I've never had any users asking me 
for Fedora/CentOS/etc packages. Granted, my user base is fairly 
technical and consists of mostly sysadmins and developers, and a large 
portion of them are on MacOS X, so I don't know whether this applies in 
general.

I don't think that Autopackage in its current form is the way to go 
forward. For a system to be successful, it must not only work in theory, 
but it must also be accepted by developers. I'd argue that a system that 
is less technically sound, but is easier to use by developers/packagers, 
has a better chance of succeeding.


But before proposing any kind of solution, I'd like you all to consider 
this question:

   What problem are we trying to solve?

A possible answer might be: "People switching from Windows can't figure 
out how to install software, and concludes that Linux sucks". However:
1. The Linux hype is long over. The Linux user base is still growing but 
I don't see a lot of Windows-to-Linux converts anymore. Most people who 
care about GUIs and usability have switched to MacOS X, leaving mostly 
the technical/hardcore guys who don't care about this kind of things to 
Linux.
2. I'm noticing that pretty much everybody these days *expects* that 
software is installed through their native package managers. I have to 
conclude that there is no problem as far as most existing Linux users 
are concerned. The people who complain about installation on Linux are 
typically not Linux users, and I don't think they'll come back even if 
we address their complaints.

Another possible answer might be: "As a developer, I don't want to 
invest manpower into creating a package for each distro".

I have the feeling that Autopackage has historically been too focused on 
the former answer than the latter answer. Everything is very end-user 
focused, but not enough developer-focused. Even though developers are 
arguably more important, because without them we can't get anything done.

---------------------------------------------------------------------
To unsubscribe, e-mail: autopackage-dev-unsubscribe-OfajU3CKLf1/[email protected]
For additional commands, e-mail: autopackage-dev-help-OfajU3CKLf1/[email protected]