This is my JSON parser. I want some practice making parsers, and JSON seemed relatively easy and useful.

API

The API is documented in json.h, but here's a quick example:

/* Compile with -DJSON_WITH_STDIO */

FILE *file = ...;
char buf[BUFSIZ];
json_reader rdr;
struct json_item item;

json_alloc(&rdr, NULL, 8, malloc, free, realloc);
json_source_file(&rdr, buf, sizeof(buf), file);
do {
	if (json_read_item(&rdr, &item) < 0) {
		/* Handle the error, the type of which is in item.type. */
	} else {
		/* Append this item using item.type, item.key, and item.val. */
	}
} while (item.type != JSON_EMPTY);

Here's the same example using zlib GZip decompression:

#include <zlib.h>

static int refill_gz(char **buf, size_t *size, void *ctx)
{
	gzFile file = ctx;
	int got = gzread(file, *buf, *size);
	if (got < 0) return -JSON_ERROR_IO;
	if ((size_t)got < *size) {
		*size = got;
		return 0;
	}
	return 1;
}

gzFile file = ...;
char buf[BUFSIZ];
json_reader rdr;
struct json_item item;

json_alloc(&rdr, NULL, 8, malloc, free, realloc);
json_source(&rdr, buf, sizeof(buf), file, refill_gz);
do {
	if (json_read_item(&rdr, &item) < 0) {
		/* Handle the error, the type of which is in item.type. */
	} else {
		/* Append this item using item.type, item.key, and item.val. */
	}
} while (item.type != JSON_EMPTY);

As you can see, all it takes is changing a callback to add another stage to the pipeline.

For a more in-depth example, see parse.c.

This library reads and writes UTF-8, but you can easily adapt it to other encodings if you wish because of the flexibility of the input source.

Tests

I got the tests from this repository, which can be found under the tests directory here, or the test_parsing directory in the original project. Many thanks to Nicolas Seriot for providing those tests publicly.

Test Failures

The currently failing tests do not seem very important to me to fix. If the user wants this parser to be compliant, it is easy to keep it that way. Just parse only one map/array and check that the input source is empty afterward.

Dependencies

Currently the library depends on the standard library for string manipulation and character classification. It depends on the standard math library for the pow function. It does not require the user to use the standard library allocation, although that is the easiest method.

The tests require Perl 5 or so to run, and I haven't tested the scripts on non-unix systems. run-tests specifically makes heavy use of POSIX system calls.