I've currently *ahem* negatively affected the efficacy of the SqueakTrunk and ReleaseSqueakTrunk builds. I'm hoping some kind soul will quickly reply to my earlier mail about cleanup, telling me what very important step I've foolishly left out: that ought to fix the problem. (And I'll turn it into a SmalltalkImage method that we can repeatedly use.)
Otherwise, on a more positive note, I've * reimplemented the entire build process in Ruby (a far more sane language than shell), * made SqueakTrunk builds off a 4.5 image, drastically reducing the build time, * broken SqueakTrunkPerformance out as a separate job, with the aim of making debugging the current problem easier, * will make ReleaseSqueakTrunk use Ruby (once we're done with the current issue), and will maintain the illusion that nothing has changed, while SqueakTrunk's artifact continues to shrink.
Lastly, ExternalPackages builds a base image off a fully updated 4.5 image and runs, each in a clean image, the test suites (where present) of AndreasSystemProfiler, Control, FFI, Fuel, Nebraska, Nutcracker, OSProcess, Quaternion, Phexample, RoelTyper, SqueakCheck, Universes, WebClient, XML-Parser, Xtreams and Zippers. And other jobs do the same with 4.3 and 4.4.
(I realise some of the packages will be unfamiliar: some are from Pharo, and some are libraries I've released and written about elsewhere, but talking about them here would be off-topic.)
frank