Thanks for all the good suggestions about this problem. Writing a
simple
allocator like suggested seems like a decent solution. I did try
another
approach first though: I remade my program so that there is a
specific point
during execution where I can free all dynamic allocations. After
implementing that I still had fragmentation issues, so after logging
all
remaining allocations between two runs a bunch of LUA allocations
showed up.
Since I doubt there are memory leaks in there, I assume that I do
something
wrong, but a call to lua_close() should free up all dynamic
allocations,
shouldn't it? I also tried a lua_gc(L, LUA_GCCOLLECT, 0) call before
closing
the state, but that didn't affect anything.
Thanks,
/Jens
-----Original Message-----
From: lua-bounces@bazar2.conectiva.com.br
[mailto:lua-bounces@bazar2.conectiva.com.br] On Behalf Of Stephen
Kellett
Sent: den 22 februari 2008 17:13
To: Lua list
Subject: Re: How to best solve fragmentation issues
Jens Andersson wrote:
separate heap. I'm not really looking forward to implement a separate
memory-manager, but this way the LUA allocations won't interfere
with the
rest of the memory.
Is there a better way to solve this?
You may be better off writing a C extension that provides an API that
simply allocates one large chunk of memory and then
allocates/deallocates that memory explicitly. Four functions should
do it:
Create heap
allocate from heap
deallocate from heap
Destroy heap
This removes you from the problem of having to write a full blown
memory
manager to replace the Lua one, plus the problems of making your
custom
memory manager as fast as the general purpose one (your simple cases
may
be faster, but the nasty cases will most likley be worse, or just
fail).
It also means that if your images are all the same size or you know
the
largest size of the images and how many images you'll need at any one
time you can ensure your heap has this space and also ensure that your
allocate()deallocate() functions prevent fragmentation.
I know I'm advocating explicit memory management in a GC'd world. But
for things like this its almost certainly a better bet. I've come
across
some similar cases in the Windows world (where you can force the
result
with VirtualAlloc/VirtualFree or
HeapCreate/HeapDestroy/HeapAlloc/HeapFree). The C allocator will never
give you the control so that you can guarantee the result.
Stephen
