From fahree at gmail.com Sun Jul 25 00:52:24 2010 From: fahree at gmail.com (=?ISO-8859-1?Q?Far=E9?=) Date: Sat, 24 Jul 2010 20:52:24 -0400 Subject: [xcvb-devel] standalone backend Message-ID: After many months, and in a frenzy of procrastination for more important things to do, I completed last week the concurrent process infrastructure necessary for XCVB to build things in parallel by forking live images rather than starting from scratch. The standalone build is not incremental yet, as I don't yet have a way of storing dependency metadata between builds. Interestingly, things work well in CCL, but several bugs plague the SBCL port: * a bug in one of SBCL, CFFI or IOlib causes corruption if GC happens while building with SBCL as the master. That's really bad. Workaround provided. (Allocate a large heap, GC before you build.) * some bug on non-portable feature somewhere in SBCL and/or my code causes the forker to fail to successfully open the named pipes when SBCL is the slave. I've independently been too tired / depressed / procrastinating to isolate these two SBCL issues. Sigh. Maybe someone has more courage than me with this regard. I've committed and pushed my state of the system. Note that you need my branch of CFFI and IOlib, QUUX-IOlib, and several other such dependencies. Note also that iolib and quux-iolib generate .so files, and that I don't at this point have a good deployment story for these files. Plenty of ungrateful hours of coding lie between here and such a story for the many CL implementations (or even one of them). CL is a harsh mistress. I'm not sure I like her anymore. [ Fran?ois-Ren? ?VB Rideau | Reflection&Cybernethics | http://fare.tunes.org ] What one person receives without working for, another person must work for without receiving. ? Adrian Rogers