Subj : Re: memory management in jsref
To   : Aleksey
From : Brendan Eich
Date : Thu Jul 29 2004 07:36 pm

Aleksey wrote:
> Hello,
> 
> To increase the performance (speed) of jsref on windows platform I use
> not standard memory manager (Doug Lea's Malloc,
> http://gee.cs.oswego.edu/dl/html/malloc.html). Performance speed is
> definitely improved and everything looks fine but I still have a
> slight doubts.
> 
> When I was inserting dlmalloc to the engine, I noticed that somewhere
> the engine uses malloc/free functions and in some other places it uses
> JS_malloc/JS_free which in their turn use same malloc/free.
> I was wondering if jsref follows any well-defined memory policy?

Yes -- everything is layered on one malloc/realloc/calloc/free 
implementation.  It doesn't matter whether you JS_malloc and then free, 
or malloc (with your own error check, calling JS_ReportOutOfMemory) and 
then JS_free (although that is odd, because JS_free adds little value 
(just null defense).

If you are trying to change JS_malloc and JS_free to call something 
else, that's the wrong path.

You need to *link* with a different libleamalloc.a or .so (Unix), or 
whatever, that exports the standard symbols malloc, free, etc.

/be

.