Re: [EXPERIMENT] Emacs with the SpiderMonkey garbage collector

[Stefan Monnier]

>> But it's not like it doesn't have downsides.
> ... and then you stopped writing! Just when you were getting to the good part!

Off the top of my head:
- `sxhash-eq` is a lot more painful to implement.
- during GC you need *double* the heap space.  If you only GC a sub-part
  of your heap (e.g. just one generation), then you only need double
  that sub-part.  So if you only use a copy-GC for the youngest generation
  it's not a big deal, but if you use stop&copy over your whole heap,
  we're talking about a 100% overhead on memory usage.
- interaction with C code is significantly more tricky since you really
  need to find *all* the pointers into your objects, including random
  transient ones that aren't Lisp_Object.
  Often, the only sane way out is to disallow access to the actual
  pointers from C: e.g. the C code only manipulates "handles" into an
  array of actual pointers.
- the most obvious concurrent-GC for stop&copy requires a read-barrier
  rather than a write-barrier.

But I do recommend taking a look at the TI Explorer II's garbage
collector, described in the article below.  It's making a really neat
use of the read-barrier!


        Stefan


@Article{Courts88,
  author =       {Robert Courts},
  title =        {Improving locality of reference in garbage-collecting
                  memory management system},
  journal =      CACM,
  year =         1988,
  volume =       31,
  number =       9,
  pages =        {1128-1138},
  doi =          {http://doi.acm.org/10.1145/48529.48536},
  xnote =        {TI Explorer II's GC!!},
  abstract =     {Modern Lisp systems make heavy use of a garbage-collecting
                  style of memory management.  Generally, the locality of
                  reference in garbage-collected systems has been very poor.
                  In virtual memory systems, this poor locality of reference
                  generally causes a large amount of wasted time waiting on
                  page faults or uses excessively large amounts of main
                  memory.  An adaptive memory management algorithm,
                  described in this article, allows substantial improvement
                  in locality of reference.  Performance measurements
                  indicate that page-wait time typically is reduced
                  by a factor of four with constant memory size and disk
                  technology.  Alternately, the size of memory typically can
                  be reduced by a factor of two with constant performance.}
}