After Matthews patch to scan quicklier through memory I<br>ran into another bug which took some time to find. The FreeBSD<br>compiler environment doesn't define __WORDSIZE, so <br>GC_MODEL_NATIVE32 was defined as a result. I patched it
<br>to check for __LP64__ as well, but I think there is something<br>smarter to discriminate by.<br><br>With that change, I got a GC_MODEL_NATIVE64 and no more<br>GC invariant assertion failures ;)<br><br>The built MLton is now happily chewing on the regression
<br>suite. It has done about 40 or something and no errors has<br>been reported yet.<br><br>I will cook up a possible patch when it is done with the regressions.<br><br>