Re: Question about Lua 5.1.3 and realloc() on life.lua

[Mike Pall <mikelu-0805@...>]

Bogdan Marinescu wrote:
> >  Umm, reinventing the wheel? And why do so without studying the
> >  literature? There's plenty out there and it will tell you that
> >  best-fit is rarely a good choice. While it has the least amount of
> >  internal fragmentation it causes the highest amount of external
> >  fragmentation and has unacceptable worst-case behaviour.
> By the way, best-fit is probably the best choice for such a simple
> allocator. As Doug Lea himself suggests:
> 
> "As shown by Wilson et al, best-fit schemes (of various kinds and
> approximations) tend to produce the least fragmentation on real loads
> compared to other general approaches such as first-fit."

Severe case of terminology mixup. I was referring to 'sequential
best-fit'. Given the context, the quote probably refers to
'segregated fit' (derived from best-fit) since dlmalloc is a
deferred coalescing segregated fit allocator.

I suggest to follow the taxonomy in:
  http://citeseer.ist.psu.edu/wilson95dynamic.html

--Mike