It's amazing how many people get this wrong.
Here's an example of a tree which the naive solution fails to reject:
5
/ \
/ \
4 6
/ \ / \
1 7 1 7
Every invocation of a naive check will succeed, since every parent is between its children. Yet, it is clearly not a well-formed binary search tree.
Here's a quick solution:
bool test(Tree* n,
int min=std::numeric_limits<int>::min(),
int max=std::numeric_limits<int>::max()) {
return !n || (
min < n->data && n->data < max
&& test(n->left, min, n->data)
&& test(n->right, n->data, max));
}
This isn't perfect, because it requires that neither INT_MIN nor INT_MAX be present in the tree. Often, BST nodes are ordered by <= instead of <, and making that change would only reserve one value instead of two. Fixing the whole thing is left as an exercise.
Here's a demonstration of how the naive test gets it wrong:
#include <iostream>
#include <limits>
#define T new_tree
struct Tree{
Tree* left;
int data;
Tree* right;
};
Tree* T(int v) { return new Tree{0, v, 0}; }
Tree* T(Tree* l, int v, Tree* r) { return new Tree{l, v, r}; }
bool naive_test(Tree* n) {
return n == 0 || ((n->left == 0 || n->data > n->left->data)
&& (n->right == 0 || n->data < n->right->data)
&& naive_test(n->left) && naive_test(n->right));
}
bool test(Tree* n,
int min=std::numeric_limits<int>::min(),
int max=std::numeric_limits<int>::max()) {
return !n || (
min < n->data && n->data < max
&& test(n->left, min, n->data)
&& test(n->right, n->data, max));
}
const char* goodbad(bool b) { return b ? "good" : "bad"; }
int main(int argc, char**argv) {
auto t = T( T( T(1),4,T(7)), 5, T(T(1),6,T(7)));
std::cerr << "Naive test says " << goodbad(naive_test(t))
<< "; Test says " << goodbad(test(t)) << std::endl;
return 0;
}