Options are good, but one algorithm that tries to get it all right is probably asking for trouble. Dolphin has an elaborate stripper, but I found that slicing out unreferenced packages would typically give me a 4 MB or so executable. I was able to strip further, but it took time and raised the possibility of runtime failures. Which tests drive the stripping? They could easily be broader than strictly necessary, causing otherwise unneeded things to be retained. I agree with Stef that good packaging is the place to start. ________________________________________ From: pharo-project-bounces@lists.gforge.inria.fr [pharo-project-bounces@lists.gforge.inria.fr] On Behalf Of Markus Fritsche [fritsche.markus@googlemail.com] Sent: Sunday, May 29, 2011 4:57 PM To: Pharo-project@lists.gforge.inria.fr Subject: Re: [Pharo-project] A radical proposal (to cut down dead code) 2011/5/29 askoh <askoh@askoh.com>:
Is Spoon trying to do this?
AFAIK, yes. As far as I understand, Craig is using a modified VM to identify sent messages and to automagically mirror them from a master to a slave image; making testcases very crucial in the process. To make a tailored image, you run all the testcases and the "needed" behavior automagically gets pushed/ pulled into the slave image. Save the slave, and you can run it indepentently. "Spoon" way - mark each sent message (within the vm) - execute a full test cycle - see what has been marked- "Tinkering way" - modify all CompiledMethods to send a "self registerAsSent: self selector object: self" by code and run a full test Two ways I could think off right now, not really deep into it.