Hi,
This is probably a simple question. I'm working on a simple numeric
vector class, and I'd like to be able to create 'views' of these
vectors which don't require making a copy of the original vector, aka:
a = numvector.new(100) -- initialize a vector w/ 100 elements, zeroed out
b = a:view(4,10) -- create a view which shares the same memory block
as 'a'
b[1] = 42 -- a sees this change: so now a[4]==42
On the C side, I have this:
typedef struct NumVector {
unsigned int start, stop;
lua_Number *values;
} NumVector;
So, in creating a new 'view', the newly created object will have the
same values array, but with possibly different values for 'start' and
'stop'.
If I were to just allocate a NumVector with a block, like:
/* unsigned int size = something */
NumVector *a = (NumVector *)lua_newuserdata(L, sizeof(NumVector) +
size*sizeof(lua_Number));
...then 'a' would be subject to garbage collection by Lua. Great!
Except that when a becomes garbage, its values array will be
deallocated, so any views that point to it will now be invalid. I
could allocate the values array separately:
NumVector *a = (NumVector *)lua_newuserdata(L, sizeof(NumVector));
a->values = (lua_Number *)lua_newuserdata(L, size*sizeof(lua_Number));
but I don't think this is correct. I'd imagine Lua could immediately
collect a->values since nothing on the Lua side will point to it. (or
does the mark and sweep follow references on the C side, too?). What
would be the cannonical way to handle this situation?
-Paul
