json_object_get_boolean() returns wrong result for objects and arrays

[plan44]

Describtion of the bug
The current implementation of json_object_get_boolean(jobj) returns 0 (false) when jobj is a json_type_object or json_type_array.
This seems to be in contrast to the expected behaviour as documented, which is:

The type is coerced to a json_bool if the passed object is not a json_bool.
integer and double objects will return 0 if there value is zero
or 1 otherwise. If the passed object is a string it will return
1 if it has a non zero length. If any other object type is passed
1 will be returned if the object is not NULL.

Steps To Reproduce

#include <json-c/json.h>
#include <stdio.h>

int main()
{
  struct json_object* jobj = json_object_new_object();
  int ob = json_object_get_boolean(jobj);
  printf("bool of a json_type_object = %d (should be 1)\n", ob);

  struct json_object* jarr = json_object_new_array();
  int ab = json_object_get_boolean(jarr);
  printf("bool of a json_type_array = %d (should be 1)\n", ab);
}

Output of this is:

bool of a json_type_object = 0 (should be 1)
bool of a json_type_array = 0 (should be 1)

Version and Platform

• json-c version: json-c-0.15 (installed via brew)
• OS: macOS 10.15.5

[hawicz]

This code has always returned 0 for objects and arrays, so I'd call this a doc bug, not a code bug. Thanks for the pulll request, but I won't be merging it currently. Instead, I added an item to the proposed 1.0 changes to consider it there.
I updated the doc and added some tests in commit 2b439ea.

[plan44]

Ok, I understand the decision, given json-c has always done it that way. Now that it is documented, I'm fine ;-)

But I still support a 1.0 change that would make this to work like the documentation said until now - because that's exactly what JS's Boolean() does and thus is widely expected behaviour.

I stumbled over this after hours of debugging a case where a message could be false or a json object, and the test was just json_object_get_boolean(), which would have worked according to the docs... (Once found, it was easy to add a wrapper which tests in the way JS would)