My observations for issue 12506 - I can reproduce it on linux - segfault happens in primtiveNextInstance - tests A-L worked in 30663 - tests A-L segfault since 30664 - doesn't segfaults anymore if you open the image, change some code, save and close the image - it segfaults if you just run a subset of the A-L tests (pharo --nodisplay Pharo.image test --junit-xml-output 'FileSystem-Tests-Core' 'FreeTypeTests' 'FileSystem-Tests-Disk' 'FuelTests' 'FileSystem-Tests-Memory' 'KernelTests') - this "fail"-subset is different for the different images (30664, 30665, ...) - it does not segfault if you omit the junit-xml-output option And someone knows what happened with the image between version 30663 and 30666, because only 30664 and 30665 need the PharoV20.sources in the same directory as the VM, for all other versions it suffices if the sources are where the image file is. (Same as in http://forum.world.st/pharo-3-0-image-td4732890.html) Nicolai