[Live-demo] Rethinking osgeo-live

[Alex Mandel]

On 10/24/2012 03:08 PM, Barry Rowlingson wrote:
> I've not really thought this through, but I'll put it out there for
> discussion...
> 
> Would the effort spent on creating the osgeo-live disc be better spent
> creating a 'portable' set of compiled applications, for Linux, Mac,
> and Windows platforms?
> 
> The advantage would be that a user wanting to try something out would
> just copy the bits they wanted to their PC, and use their existing
> operating system. An installer would just copy what the user wanted
> and handle dependencies (much like osgeo4w, but multi-platform).
> 
I'm not actually sure the current team has that expertise. I've never
gotten anything to compile correctly on windows (part of why I don't use
it) and I'm still struggling to properly package for debian/ubuntu.
I'm sure you're aware of the GIS on a stick that Jo works on
occasionally. There's also been efforts to do osgeo4w on a stick.

> It would also be more likely to be usable on machines with a
> locked-down boot sequence. Our central IT people supply desktops with
> passworded BIOS settings and HD-before-DVD boot sequence. They also
> physically lock the cases. Killjoys.
> 
I use Virtual Machines in my labs, still requires admin to get Vbox
installed, but the lab management has given me that much. Almost anyone
can find and old discarded laptop to try in the corner though, and when
they see what they like approach IT to help them install what they need.

> There would also be no need for admin privileges - something that
> blights system package management systems like apt and rpm.
> 
> Is it also true that in the very near future PCs will have some kind
> of trusted boot system? So that alternate operating systems would
> require signing (or I read something about a signed mini-bootloader
> being developed to possibly get round this...). Will that effect live
> DVD boots? That could be a pain...
> 

Canonical is buying a signature for Ubuntu(and derivatives), and the
Linux Foundation I think is getting a blanket one for linux in general.

> The disadvantages:
> 
>  Loss of total control - we wouldn't know exactly what OS the programs
> were going onto, so documentation might look wrong.
> 
>  Binary compatibility - how can you ensure your binaries work with
> assorted Linux versions? That might be the show-stopper here, although
> I'm pretty sure I've recently installed Linux software from
> one-size-fits-all binaries. Would the only compatibility be to do with
> libc and the kernel? I envisage practically everything being on there,
> including things like the Qt library and a JVM.
>
Oh static linked binaries... spatialite does this and it works most
places but is the ancient archaic windows way of doing things.

>  There would be a need to re-tool all the osgeo-live development
> chain, and write an installer.
> 
>  There would be three versions - Linux, Windows, Mac (or four if
> anyone wants to compile for OpenSolaris...)
> 
> I just think there's greater longevity and value in the distribution
> of a collection of ready-to-run, run-anywhere packages than a live
> boot disk these days.
> 

One of my goals all along has been to simply encourage packaging for
major OS. Hence if you make an ubuntu deb we put it on, if you've got a
win or mac installer we try to include.

Java, .Net, etc all attempted to solve the write once run anywhere and
it's never really worked. If we really wanted to push that notion -
browser based apps is the way to go these days, js+html5 or a VM running
in a browser (eg. Tilemill is a nodejs app).

> Thoughts?
> 
> Barry

In review, I'm still in favor of every 6 months. We are really close to
full automation of the entire build process on a timer (right now a
person has to trigger 1-3 steps). Once that's in place we can start
showing people how to build a controller file to generate custom builds
which is where I think many want to go next.


Enjoy,
Alex