Jansson is thread safe and has no mutable global state. The only exceptions are the hash function seed and memory allocation functions, see below.
There’s no locking performed inside Jansson’s code, so a multithreaded program must perform its own locking if JSON values are shared by multiple threads. Jansson’s reference counting semantics may make this a bit harder than it seems, as it’s possible to have a reference to a value that’s also stored inside a list or object. Modifying the container (adding or removing values) may trigger concurrent access to such values, as containers manage the reference count of their contained values. Bugs involving concurrent incrementing or decrementing of deference counts may be hard to track.
The encoding functions (json_dumps() and friends) track reference loops by modifying the internal state of objects and arrays. For this reason, encoding functions must not be run on the same JSON values in two separate threads at the same time. As already noted above, be especially careful if two arrays or objects share their contained values with another array or object.
If you want to make sure that two JSON value hierarchies do not contain shared values, use json_deep_copy() to make copies.
To prevent an attacker from intentionally causing large JSON objects with specially crafted keys to perform very slow, the hash function used by Jansson is randomized using a seed value. The seed is automatically generated on the first explicit or implicit call to json_object(), if json_object_seed() has not been called beforehand.
The seed is generated by using operating system’s entropy sources if they are available (/dev/urandom, CryptGenRandom()). The initialization is done in as thread safe manner as possible, by using architecture specific lockless operations if provided by the platform or the compiler.
If you’re using threads, it’s recommended to autoseed the hashtable explicitly before spawning any threads by calling json_object_seed(0) , especially if you’re unsure whether the initialization is thread safe on your platform.
Memory allocation functions should be set at most once, and only on program startup. See Custom Memory Allocation.
Jansson works fine under any locale.
However, if the host program is multithreaded and uses setlocale() to switch the locale in one thread while Jansson is currently encoding or decoding JSON data in another thread, the result may be wrong or the program may even crash.
Jansson uses locale specific functions for certain string conversions in the encoder and decoder, and then converts the locale specific values to/from the JSON representation. This fails if the locale changes between the string conversion and the locale-to-JSON conversion. This can only happen in multithreaded programs that use setlocale(), because setlocale() switches the locale for all running threads, not only the thread that calls setlocale().
If your program uses setlocale() as described above, consider using the thread-safe uselocale() instead.