On Thu, Dec 11, 2008 at 11:37 AM, Gerard Meijssen <gerard.meijssen(a)gmail.com
wrote:
Hoi,
Last year Nikerabbit was enrolled in a Finnish Summer of Code project. He
did a ton of great work for Betawiki as part of this project. The Liquid
Threads project was a GSOC project. It is used by the WikiEducator project
and as such I would rate it successful.
That's great, but neither Betawiki nor WikiEducator are the WMF, nor
are those functionalities being used (beyond the localizations provided
from BW) within the WMF. That's precisely what this is about, making
use of the GSOC-style of development for MediaWiki itself.
When you look at our own bigger projects, SUL took a crazy amount of time to
materialise, we are still not able to produce
predictable data dumps. When
you look at commercial projects, at least 50% of such projects fail to meet
expectations. The notion that classical "in the office" projects do better
is not one I share.
SUL required a massive amount of work to coordinate, not to mention the
task of coming up with a model that not only works, but is scalable to the
WMF's needs and also actually make sense. Good data models are essential.
Dumps are another thing that requires careful work and coordination. Poor
execution leads to poor results. It's bad enough having someone e-mail the
list(s) once every month or two saying "new dumps please," but imagine if
we provided dumps that were just inherently bad.
I fail to see where you get this 50% of commercial projects fail. Without a
reliable source, I have to assume you've just made up this number and have
never worked in software development.
When we are to do a proper job for summer of code projects, obviously all
our existing developers are most likely to do the
better job. Nikerabbit's
project is a case in point. If there are observations why such a project
does not work out as well as we hope, we should address those issues. The
most important thing achieved with a summer of code project is not only the
software but also the experience given to what we hope will be a developer
who stays with our project after his project.
Thanks,
GerardM
Tim never said that GSOC is a bad model. Nor did he say that it produces bad
results. He simply said that based on previous experiences, it hasn't worked
_for us_. And it's true. Looking at last summer's projects, I don't see a
lot of
results:
1) HTMLDiff - ended up as a highly experimental, still somewhat buggy,
disabled-
by-default feature that was an i18n mess (and still might not be 100% fixed)
2) Category Redirects - what ever happened to this? Was this ever merged
from
branch to core? The branch hasn't been touched since August, at the least.
A model failing in one use case doesn't indicate a failed model overall; it
simply
means it doesn't work in this situation. :)
-Chad