Fri Nov 13 13:24:17 CET 2009

GC restarts: disallowed by default.

FIXME: bunch of (all?) _sc_make_aref() calls might be restarted!  it's
probably simplest to probe the GC free space inside the interpreter
loop to at least make those calls succeed.

In general: define a global nb of cells that is _guaranteed_ to be
available to all primitives invoked from _sc_step().  This might be
extended to a mechanism that completely prevents GC restarts inside
all primitives, turning it into an error.

Primitives that allocate a lot of cells then can be the exception:
they could allow GC restart manually.