Ben what we want is the following - we have different distributions for pharo1.0, 1.2.14 and automatically we run tests and rules checking systematically - when a package has too many flagged items -> goes into the inbox of the distributions - Package should embed their false positive to SmallLint so that we can run them for real. - when people propose fixes we want the system to automatically load and test. This is what Ulysse the Monkey is already doing and this is great
Stef
Just a very general view... that in paid positions, you don't always get to work on the sexy stuff. One of the downsides of open-source is the chance that developers might focus on the fun 80% of their ideas before moving on to the next-big-thing(c). It is easy to develop something to be used by oneself. It is much harder to develop something to be used by others - particularly in providing a sense of how to use the system by way of well documented examples. There is a lot of drudge work to make a system successful and stable. Knowing there is a paid position that might deal with part of that work definitely increases confidence in the system. One new idea. I really like the way Wordpress manages their thousands of Plugins with the "Compatability" sidebox where you choose a Wordpress version and a Plugin version and it indicates how compatible they are based on user votes. However rather than (or as well as) user voting I think that TestRunner results from applications could be used - and also presented in a tabulated form rather than just pull-down choices showing a single sample point. A continuous integration and testing system would take combinatorial approach to testing registered Metacello configurations - which might take a lot of processing power and leads to...
A second idea for a distributed continuous integration test system. I would personally be happy to contribute cpu cycles to running a Pharo test system on top of a virtual machine, which from a central server obtains a random set of Metacello configurations to install, runs tests and reports results to the central server. Apart from gaining the processing power to possibly test all permutations of configurations, it also provides a way for new community members to feel like they are contributing, which preceeds them getting more involved. As an example in line with my first comment, perhaps the initial proof of concept would be done as a research project, and then the paid professional would polish off the stability and usability of it.
Yes this is a neat idea :) To contribute you can already load a fixe and tell us if the tests are ok. Or just if the fix is loading :). Stef
cheers, Ben