I've committed a patch that currently fixes the t/dynclasses/*.t errors
with gcc 3.3.3 and possibly others.
While I wasn't able to fully track down the real problem, it seems that
dynamic library loading doesn't stand a DOD run, which is artificially
triggered by string_compare, when --gc-debug is on.
The whole static MMD table stuff, where the error hit the surface, is
doomed anyway, and eventually we have to create all persistent core
structures in constant (non-GCed) memory pools.
leo