[sldev] Webkit versus Second Life

[Carlo Wood]

On Fri, Mar 13, 2009 at 08:14:28AM -0500, Soft wrote:
> Watching an Apple Safari/Webkit developer intro talk, they did
> something pretty cool: They demonstrated that you can fetch and build
> Webkit using nothing but your mouse. They assume you have the
> developer tools and subversion installed. The rest is su-and-say
> enough that you can just copy and paste three lines from the browser.
> One to do the svn fetch, one to build, and one to launch the built
> app.

That is bullshit; each pasted command can start a plethora of other
commands. I could write a script that downloads, build, installs
AND starts SecondLife - including all shared libraries if you
wish and call it "we_are_so_cool". Then you "beat" Apple Safari/Webkit:
just ONE line to paste!

> We're pretty far away from that. We rely on a handful of libraries
> that we can't provide on our own. We need some extra development tools
> installed. We have library and artwork bundles apart from the source
> bundle. We have separate steps for configuring and for building.

I consider most of that a plus! Having a "black box" means less
flexibility and ONLY works in a FULLY (version) controlled environment.
Microsoft and Apple already (attempt to) do that (and still fail
often -- except during a presentation). Linux has chosen
a different path. I strongly believe that a non-black box, and
flexibility is worth more than a "one click" process that either
works or not (that being said, I AM author of a few auto-magic
scripts like http://www.xs4all.nl/~carlo17/howto/build-nvidia-kernel
and http://www.xs4all.nl/~carlo17/vvinstall2
Look at those scripts to have an idea how "cool/good" one-click
installations are... Do you really want this? Nevertheless, I wrote
those for end-users that otherwise would simply have nothing at all.
Developers should be able to understand each step. The more steps
the better. What we NEED is a HOWTO that explains every step in
detail (and thus, many copy&paste lines). Most of my HOWTO's are
that way, look for example at http://www.xs4all.nl/~carlo17/howto/encode.html
or http://www.xs4all.nl/~carlo17/svn/index.html with many
little copy&paste blocks instead of one-script-fits-all.

> An awful lot of new devs get lost somewhere in the process.
> 
> Of all the above, if we were going to focus on knocking out just one
> step next, which do you think would be most valuable? Which is the
> highest hurdle?

A new developer has NO value unless he has an overview of the process,
the components involved and how everything hangs together.
New developers need to learn a lot of things. The best way to
learn is by doing. Changing a whole install/compile step to
a one-click won't make them any wiser and they will be of no
value imho.

Nevertheless-- I can answer your question ;). The answer is
*precisely* what the Open Metaverse Viewer project's goal is:
* Only use opensource components and to remove any non-free dependencies.

Your next step should be get rid of any closed source third-party
libraries/software. Then we need a webpage that outlines each
step for new developers to follow, rather than one powerful
script. We (OMV developers) ran into this too, so you can have
a look at our first attempts to outline things that need to be
done on http://omvviewer.byteme.org.uk/source.shtml
(courtesy Robin Cornelius) and in more detail
http://omvviewer.byteme.org.uk/git.shtml

> Another cool thing they do is making sure the nightly builds are
> completely stand-alone. Testing a build is as simple as unpacking a
> zip file and running a file in the contents - no installers permitted,
> no writing to files outside that directory. The idea is that it's
> possible to keep an array of nightlies and binary chop through
> versions to find regressions. Mac and Linux are pretty much at this
> stage. Would this be preferable for Windows BSI nightlies?

-- 
Carlo Wood <carlo at alinoe.com>