>> The main cost of a concurrent GC is that the application code has to be
>> changed to cooperate with the GC.

> Frankly, I don't think so.  Lorenz Huelsbergen and Phil Winterbottom
> implemented their VCGC algorithm [1] for the Limbo programming language
> under Inferno and for SMLNJ as well, and their use of that concurrent
> garbage collection algorithm required no changes to any Limbo or SMLNJ
> application code.

I meant *compiled* application code, of course.  I.e. you can't change the
GC after compilation.  Admittedly, the VCGC algorithm is sufficiently
non-intrusive on the mutator that it might be OK to pay its cost even
when not using the VCGC algorithm.


        Stefan