Closed DaDuFan closed 4 years ago
Hi! Thanks for reporting this.
I am afraid this issue has no simple and quick solution. It's because there is a nature of JSON where keys should be represented as quoted strings. And it's a big question to me is it good for json-dto to hide that nature from a user. We have do discuss this issue inside our team.
As a workaround, I can propose to use a special wrapper for non-string keys. Something like:
template<typename T>
struct key_wrapper {
T v_;
... // Constructors
bool operator==(const key_wrapper & other) const noexcept { return v_ == other.v_; }
bool operator<(const key_wrapper & other) const noexcept { return v_ < other.v_; }
...
operator T() const { return v_; }
...
};
template<typename T>
void read_json_value(
key_wrapper<T> & value,
const rapidjson::Value & from)
{
... // Read a string and convert it to T.
}
template<typename T>
void write_json_value(
const key_wrapper<T> & value,
rapidjson::Value & object,
rapidjson::MemoryPoolAllocator<> & allocator)
{
... // Convert value into a string and store it into object.
}
In that case, you can use std::map<key_wrapper<int>, int>
instead of std::map<int, int>
.
Hi, @DaDuFan!
There is the following idea.
We can add a couple of new types to json-dto:
template<typename T>
struct const_associative_container_key {
const T & v;
};
template<typename T>
struct mutable_associative_container_key {
T & v;
};
There will be a couple of new versions of write_json_value
and read_json_value
for those new types:
template<typename T>
void write_json_value(
const_associative_container_key<T> what,
rapidjson::Value & object,
rapidjson::MemoryPoolAllocator<> & allocator)
{
// Just use write_json_value for T.
write_json_value(what.v, object, allocator);
}
template<typename T>
void read_json_value(
mutable_associative_container_key<T> what,
const rapidjson::Value & from)
{
// Just use write_json_value for T.
read_json_value(what.v, from);
}
The code of read_json_value
/write_json_value
for associative containers will be modified this way:
--- read_json_value( key, it->name );
+++ read_json_value( mutable_associative_container_key{key}, it->name );
read_json_value( value, it->value );
--- write_json_value( kv.first, key, allocator );
+++ write_json_value( const_associative_container_key{kv.first}, key, allocator );
write_json_value( kv.second, value, allocator );
So in such an approach, you will have a possibility to place new overloads into json_dto
namespace:
// Somewhere in your source code:
namespace json_dto {
inline void write_json_value(
const_associative_container_key<int> what,
rapidjson::Value & object,
rapidjson::MemoryPoolAllocator<> & allocator)
{
write_json_value(std::to_string(what.v), object, allocator);
}
inline void read_json_value(
mutable_associative_container_key<int> what,
const rapidjson::Value & from)
{
std::string str_v;
read_json_value(v, from);
what.v = std::atoi(str_v.c_str());
}
}
If you like that approach we can update json-dto in a couple of days.
Hi @eao197 ,
sorry for my late reply, and thanks for your fast and constructive reply.
Your idea with the new types sound great, and I think it would be a great improvement for json_dto. It is non intrusive and allows to easily handle int-keys.
If you implement it, I can guarantee that at least I will use it :-)
It seems there is another flaw in json-dto support for custom formatters (introduced in 0.2.10 in the form of Reader_Writers): in the case of custom Reader_Writer and container-like fields the Reader_Writer will be applied to the whole container, not to items of the container.
I hope to address both of those issues in one json-dto's update. And now I'm searching for a way to do that :)
Hi, @DaDuFan !
An updated version of json-dto is in the master branch. Changes are described in README.
There are also two new examples. One shows the simplest approach with overloading read/write_json_value
functions. Another shows more complex case where two std::map<uint32_t, int>
have to be represented differenly.
If there won't be any surprises I'll make a release of v.0.2.11 tomorrow.
Sounds great. Thank you very much!
I will start using it in the next days.
Hi @eao197 ,
just wanted to say thanks, again!
I'm using the release v.0.2.11 now.
First, thanks for this great library.
The implementation of map_like_associative_container does work only with string-keys, so a std::map with an int key fails at runtime, because object.AddMember( key, value, allocator ); asserts that key is a string.
e.g.:
As I needed a fast solution, I made a quick hack using std::to_string
https://github.com/DaDuFan/json_dto/commit/0158c4a359afb09cc7bb3ece312f04731d055e64
This hack is not good enough to be published, therefore I made no Pull request, but maybe you can use it as an inspiration for a better solution.