Re: [Loki-lib-general] Proposed changes to Loki's Small-Object Allocator

[<syn...@gm...>]

Richard Sposato wrote:
 >> I've read your proposal and here are some points:
 >> - I don't understand why you get trouble with the compiler when
 >> implementing
 >>   new [] and delete []. If you implement it by your own you can manage
 >>   everything by your own.
 >>   Or is the problem that when deallocating a array the information which
 >>   version of new[] was used is lost?
 >
 >
 > It is actually both problems.  I can't quite manage everything on my own
 > with the current array allocation implementation.  And some information
 > is lost between the array allocation and the array deallocation.  I try
 > to store off some info just before the array overhead, but that does not
 > always work.
 >
 > When I test array allocation, all allocation and deallocation requests
 > were passed through the small-object allocator, even those that were too
 > large for it.  Some parts of my implementation deal with determining if
 > a deallocation request should be handled by the small-object allocator,
 > or by the default allocator.
 >

What are the possible solutions?

1. store in the array the information which new[] was used
2. remember how each object was created somewere else
1. don't use the default allocator, implement new[] also for large objects

Comments:
1. - you've tried this, but it only works with msvc.
    - use different implementations for different compilers, but
      it's hard to maintain such code.

2. - it's memory intensive
    - it needs additional time to search the vector/list of remembered pointers
    + maybe the cases of calling global new[] is very seldom

3. - work must be done on a subject which is already solved by the global new.


Peter