nson

nson is a data framework for C with a very fast JSON and property list parser.

To run the tests suite:

make check

To run the benchmarks with other parsers:

make speed

for the benchmark the following libraries and headerfiles are needed:

• proplib
• libucl
• json-c
• jansson

To aquire the bench file wget is used. For converting it into plist, node and npm need to be installed.

To setup the environment in VoidLinux run the following command:

xbps-install base-devel proplib-devel libucl-devel json-c-devel jansson-devel

Goals

• Bootstrapable - NSON should be useable without fancy dependencies or build systems. That makes it possible to use it during bootstrapping where only very few base dependencies are available. (gcc, libc, make)
• Fast - NSON should be fast to setup. It is designed to be used for command line tools that won't run for long. So NSON avoids building hashmaps and uses linear/binary search which is expected to be faster in programs of short lifetime.
• mutex-/lock-less NSON provides basic thread functionality, which will be used for mapping or filtering collections of data. If you need more complex threading setups, you must lock the data by your own.

Memory Model

NSON does not rely on reference counting, garbage collection, or data locking. Instead the user is adviced to follow a few simple rules that avoid memory leaks:

• If NSON returns a pointer, NSON cares about the memory. The data is borrowed to you. Never clean borrowed data.

• If NSON expects a pointer, it will not care about freeing the memory. The data is borrowed to nson and won't be mutated after the function returns.

  • The big exception is nson_init_data(). NSON will try to free the value of the created element once nson_clean() is called on the field. This was added to have a way to save binary data in an NSON field without memcpy'ing it.

If this concept sounds familiar, yes, it is a similiar concept used in rust. :)