Segmentation fault on G++ when trying to assign json string literal to custom json type.

[lesf0]

Bug Report

This is the minimal code demonstrating the problem.

#include <iostream>
#include "json.hpp"
#include "fifo_map.hpp"

template<class K, class V, class dummy_compare, class A>
using my_workaround_fifo_map = nlohmann::fifo_map<K, V, nlohmann::fifo_map_compare<K>, A>;
using my_json = nlohmann::basic_json<my_workaround_fifo_map>;

int main()
{
	my_json foo = R"([1, 2, 3])"_json;
	
	std::cout << foo << std::endl;
}

$ g++ --version
g++ (GCC) 7.3.0
$ uname -a
Linux daddy 4.14.15-1-ARCH #1 SMP PREEMPT Tue Jan 23 21:49:25 UTC 2018 x86_64 GNU/Linux

[nlohmann]

Out of curiosity: does it work without fifo_map?

[lesf0]

Do you mean nlohmann::json instead of my_json in the code above? Yes, it works just fine.

[nlohmann]

I'll check tonight.

[gregmarr]

Do you have a stack trace? Does it happen with just the assignment line, or does it require the cout?

[lesf0]

cout here is just to suppress unused variable warning, the program actually fails on assignment. I can run this programin debugger tomorrow and attach its output here if it's required.

[gregmarr]

Just thought that if you had it handy, we might be able to see something before Niels gets to look at it tonight.

[nlohmann]

When I call my_json foo = my_json::parse("[1, 2, 3]"); instead of my_json foo = R"([1, 2, 3])"_json;, the error disappears. The problems seems to be the definition of _json which does not return basic_json, but json.

stacktrace.txt

The error can be reduced to an assignment from json to the user-defined version:

#include "json.hpp"
#include "fifo_map.hpp"

template<class K, class V, class dummy_compare, class A>
using my_workaround_fifo_map = nlohmann::fifo_map<K, V, nlohmann::fifo_map_compare<K>, A>;
using my_json = nlohmann::basic_json<my_workaround_fifo_map>;

int main()
{
    nlohmann::json x = 1;
    my_json y = x;
}

[nlohmann]

I think the problem is that the copy constructor/assignment operator/whatever thinks assigning a json object to nlohmann::basic_json<my_workaround_fifo_map> is a good idea. I think this should be avoided in the first place.

[nlohmann]

In #986 there is a fix for a related issue. Maybe it also helps here.

[nlohmann]

With PR #986, the original code example compiles and runs without issues. Output is [1,2,3].