The conversative GC still appears to give problems, even though I narrowed it down, getting that stackbase pointer is tricky! Even using the register which supposedely stores it doesn't work, even tho the crash no longer appears in collector.cpp, it now appears in value.cpp (but i still believe it has to do with corruption caused by the GC) But, testkjs now has the following results: 19 tests fixed (good!), 104 regressions found 967 test(s) selected, 962 test(s) completed, 245 failures reported (25.46% failed) Of those 245, most are really buggy JavaScriptCore completely and totally dieing. But as can be seen, its almost ready for primetime. Alot of the code is ported over, and it works really nicely as long as the GC doesn't kick in. In fact, if anyone has alot of RAM handy (my 512 ain't enough :( ) on a Wintel machine, i'll send the source to them, and let them try running it w/o the GC enabled and see if its as stable as the OS X version. Reason alot of RAM? without that GC, some tests can use alot of RAM. the biggest ive seen was 450, XP promptly BSODed when it tried to use a tiny bit more (out of physical RAM, couldnt page ANYTHING else).