Okay, this will be a pretty long digest, but hopefully worth it. You can see a testcase of the code I've presented as my conclusion here and obtain a modular version of the code here. My suggestion would be that you use a structure like this:
struct {
size_t top;
T value[];
}
The reason you probably shouldn't use classical linked lists for this (or anything, really) is covered by this video courtesy of Bjarne Stroustrup. The basis of the problem is that the majority of your overhead is in allocation and cache misses which don't occur so much when you keep everything in one allocation.
If I were to write this for convenient use, perhaps:
#define stack_of(T) struct { size_t top; T value[]; }
This should allow you to declare empty stacks fairly sensibly, like:
int main(void) {
stack_of(int) *fubar = NULL;
}
This is familiar enough to templates in other languages to work fairly well, and also not a hideous abuse of the preprocessor. I'm sure I've written a push_back
function somewhere which we can adapt to this version of push
which I've linked to externally as it's not important for the conclusion of this answer (bear with me here; we'll come back to that momentarily)...
So now we have stack_of(T)
and push(list, value)
which we can use like:
int main(void) {
stack_of(int) *fubar = NULL;
push(fubar, 42);
push(fubar, -1);
}
The simplest solution for pop
might be something like:
#define pop(list) (assert(list && list->top), list->value[--list->top]))
... but this does suffer a drawback we'll discuss later. For now we have as a testcase:
int main(void) {
stack_of(int) *fubar = NULL;
int x;
push(fubar, 42);
push(fubar, -1);
x = pop(fubar); printf("popped: %d\n", x);
x = pop(fubar); printf("popped: %d\n", x);
x = pop(fubar); printf("popped: %d\n", x);
}
... and as you'll see during debug the assert
fails during execution telling us we've popped more than we've pushed... probably a good thing to have. Still, this doesn't actually reduce the size of the stack. To do that we actually need something more like push
again, except we get rid of these lines:
list->top = y; \
list->value[x] = v; \
So there's an opportunity for refactoring. Thus I bring you operate()
:
#define operate(list, ...) { \
size_t x = list ? list->top : 0 \
, y = x + 1; \
if ((x & y) == 0) { \
void *temp = realloc(list, sizeof *list \
+ (x + y) * sizeof list->value[0]); \
if (!temp) \
return EXIT_FAILURE; \
list = temp; \
} \
__VA_ARGS__; \
}
Now we can redefine push
in terms of operate
:
#define push(list, v) operate(list, list->value[x] = v; list->top = y)
... and pop
looks kind of like it did before, but with an invocation of operate
on the end to cause list
to shrink (from quadruple its size, for example when you've popped 3 elements off of a list of 4) to no larger than double its size.
#define pop(list) (assert(list && list->top), list->value[--list->top]); \
operate(list, )
Summing it all up, you can see a testcase of the code I've presented here and obtain a modular version of the code here...