I am attempting to generate a hash value for an object that contains a boost::property_tree (boost::property_tree::basic_ptree<std::basic_string<char>, std::basic_string<char> >). Searching through the boost header files for property tree I cannot find any defined hash_value function for it. Basic example of what I am trying to achieve:
class MyClass{
public:
friend std::size_t hash_value(const MyClass & obj);
private:
boost::property_tree m_data;
}
inline std::size_t hash_value(const MyClass & obj){
std::size_t seed = 0;
boost::hash_combine(seed,obj.m_data);
return seed;
}
This code will fail to compile with: "no matching function for call to 'hash_value(const boost::property_tree &)'"
My question: Is hash_value defined for boost::property_tree in some header file I haven't found. If not, what is the boost idiomatic way to hash a property_tree by traversing it?
Should I use use ptree serialization to convert to std::string and hash that, or manually traverse the tree and create a recursive hash?
Just specialize hash<> for ptree:
#include <boost/property_tree/ptree.hpp>
#include <boost/functional/hash.hpp>
namespace boost {
template<typename Key, typename Data, typename KeyCompare>
struct hash<boost::property_tree::basic_ptree<Key, Data, KeyCompare> > {
size_t operator()(boost::property_tree::basic_ptree<Key, Data, KeyCompare> const& pt) const {
std::size_t seed = 0;
boost::hash_combine(seed, pt.template get_value<std::string>());
boost::hash_range(seed, pt.begin(), pt.end());
return seed;
}
};
}
That's enough! Here's a small MyClass that reads from a json-like literal:
#include <boost/property_tree/json_parser.hpp>
class MyClass{
public:
MyClass(std::string const& json) {
std::istringstream iss(json);
read_json(iss, m_data);
}
private:
boost::property_tree::ptree m_data;
friend inline std::size_t hash_value(const MyClass& obj){
std::size_t seed = 0;
boost::hash_combine(seed, obj.m_data);
return seed;
}
};
Now you can test it
#include <iostream>
int main() {
for (std::string const data : {
R"({"a":[1,2,3],"b":{"nest":"hello","more":"world"}})",
R"({"b":{"nest":"hello","more":"world"},"a":[1,2,3]})",
R"({ })",
R"({})",
})
{
MyClass o(data);
std::cout << "object hash: " << hash_value(o) << " " << data << "\n";
}
}
Prints:
object hash: 3573231694259656572 {"a":[1,2,3],"b":{"nest":"hello","more":"world"}}
object hash: 11176663460548092204 {"b":{"nest":"hello","more":"world"},"a":[1,2,3]}
object hash: 3864292196 { }
object hash: 3864292196 {}
For many containers, hash<> assumes a corresponding equality comparator. If they don't match, you get Undefined Behaviour.
You might be tempted to define the hash_range in terms of the ordered (associative) interface of ptree:
boost::hash_range(seed, pt.ordered_begin(), pt.not_found()); // CAUTION
This has the advantage that {"a":1,"b":2} would match {"b":2,"a":1}.
Don't do this, unless you know what you're doing. Specifically, you need to pass a compatible equality comparator to every container/algorithm that use this hash.
If you write it like this, and test using a driver like:
int main() {
MyClass a{ R"({"a":[1,2,3],"b":{"nest":"hello","more":"world"}})" },
b{R"({"b":{"nest":"hello","more":"world"},"a":[1,2,3]})" },
c{R"({ })" },
d{R"({})" };
for (auto& lhs : {a,b,c,d})
for (auto& rhs : {a,b,c,d})
{
std::cout << "hash: " << hash_value(lhs) << " " << hash_value(rhs) << " - equality: " << std::boolalpha << (lhs==rhs) << "\n";
if ((hash_value(lhs) == hash_value(rhs)) != (lhs==rhs))
std::cout << " -- MISMATCH\n";
}
}
It will print:
hash: 10737438301360613971 10737438301360613971 - equality: true
hash: 10737438301360613971 10737438301360613971 - equality: false
-- MISMATCH
hash: 10737438301360613971 3864292196 - equality: false
hash: 10737438301360613971 3864292196 - equality: false
hash: 10737438301360613971 10737438301360613971 - equality: false
-- MISMATCH
hash: 10737438301360613971 10737438301360613971 - equality: true
hash: 10737438301360613971 3864292196 - equality: false
hash: 10737438301360613971 3864292196 - equality: false
hash: 3864292196 10737438301360613971 - equality: false
hash: 3864292196 10737438301360613971 - equality: false
hash: 3864292196 3864292196 - equality: true
hash: 3864292196 3864292196 - equality: true
hash: 3864292196 10737438301360613971 - equality: false
hash: 3864292196 10737438301360613971 - equality: false
hash: 3864292196 3864292196 - equality: true
hash: 3864292196 3864292196 - equality: true
The MISMATCH warnings indicate that equality and hash do not agree.
If you run the test-driver with the original hash (above the fold) it will print:
hash: 3573231694259656572 3573231694259656572 - equality: true
hash: 3573231694259656572 11176663460548092204 - equality: false
hash: 3573231694259656572 3864292196 - equality: false
hash: 3573231694259656572 3864292196 - equality: false
hash: 11176663460548092204 3573231694259656572 - equality: false
hash: 11176663460548092204 11176663460548092204 - equality: true
hash: 11176663460548092204 3864292196 - equality: false
hash: 11176663460548092204 3864292196 - equality: false
hash: 3864292196 3573231694259656572 - equality: false
hash: 3864292196 11176663460548092204 - equality: false
hash: 3864292196 3864292196 - equality: true
hash: 3864292196 3864292196 - equality: true
hash: 3864292196 3573231694259656572 - equality: false
hash: 3864292196 11176663460548092204 - equality: false
hash: 3864292196 3864292196 - equality: true
hash: 3864292196 3864292196 - equality: true