1472 lines
		
	
	
		
			38 KiB
		
	
	
	
		
			C
		
	
	
	
	
	
			
		
		
	
	
			1472 lines
		
	
	
		
			38 KiB
		
	
	
	
		
			C
		
	
	
	
	
	
/*
 | 
						|
 * Copyright (c) 2010 Kungliga Tekniska Högskolan
 | 
						|
 * (Royal Institute of Technology, Stockholm, Sweden).
 | 
						|
 * All rights reserved.
 | 
						|
 *
 | 
						|
 * Portions Copyright (c) 2010 Apple Inc. All rights reserved.
 | 
						|
 *
 | 
						|
 * Redistribution and use in source and binary forms, with or without
 | 
						|
 * modification, are permitted provided that the following conditions
 | 
						|
 * are met:
 | 
						|
 *
 | 
						|
 * 1. Redistributions of source code must retain the above copyright
 | 
						|
 *    notice, this list of conditions and the following disclaimer.
 | 
						|
 *
 | 
						|
 * 2. Redistributions in binary form must reproduce the above copyright
 | 
						|
 *    notice, this list of conditions and the following disclaimer in the
 | 
						|
 *    documentation and/or other materials provided with the distribution.
 | 
						|
 *
 | 
						|
 * 3. Neither the name of the Institute nor the names of its contributors
 | 
						|
 *    may be used to endorse or promote products derived from this software
 | 
						|
 *    without specific prior written permission.
 | 
						|
 *
 | 
						|
 * THIS SOFTWARE IS PROVIDED BY THE INSTITUTE AND CONTRIBUTORS ``AS IS'' AND
 | 
						|
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 | 
						|
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 | 
						|
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE INSTITUTE OR CONTRIBUTORS BE LIABLE
 | 
						|
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 | 
						|
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 | 
						|
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 | 
						|
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 | 
						|
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 | 
						|
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 | 
						|
 * SUCH DAMAGE.
 | 
						|
 */
 | 
						|
 | 
						|
#include "baselocl.h"
 | 
						|
#include <ctype.h>
 | 
						|
#include <base64.h>
 | 
						|
 | 
						|
#ifndef WIN32
 | 
						|
#include <langinfo.h>
 | 
						|
#endif
 | 
						|
 | 
						|
static heim_base_once_t heim_json_once = HEIM_BASE_ONCE_INIT;
 | 
						|
static heim_string_t heim_tid_data_uuid_key = NULL;
 | 
						|
 | 
						|
static void
 | 
						|
json_init_once(void *arg)
 | 
						|
{
 | 
						|
    heim_tid_data_uuid_key = __heim_string_constant("heimdal-type-data-76d7fca2-d0da-4b20-a126-1a10f8a0eae6");
 | 
						|
}
 | 
						|
 | 
						|
struct twojson {
 | 
						|
    void *ctx;
 | 
						|
    void (*out)(void *, const char *);
 | 
						|
    size_t indent;
 | 
						|
    heim_json_flags_t flags;
 | 
						|
    int ret;
 | 
						|
    int first;
 | 
						|
};
 | 
						|
 | 
						|
struct heim_strbuf {
 | 
						|
    char *str;
 | 
						|
    size_t len;
 | 
						|
    size_t alloced;
 | 
						|
    int	enomem;
 | 
						|
    heim_json_flags_t flags;
 | 
						|
};
 | 
						|
 | 
						|
static int
 | 
						|
base2json(heim_object_t, struct twojson *, int);
 | 
						|
 | 
						|
static void
 | 
						|
indent(struct twojson *j)
 | 
						|
{
 | 
						|
    size_t i = j->indent;
 | 
						|
    if (j->flags & HEIM_JSON_F_ONE_LINE)
 | 
						|
	return;
 | 
						|
    if (j->flags & HEIM_JSON_F_INDENT2)
 | 
						|
        while (i--)
 | 
						|
            j->out(j->ctx, "  ");
 | 
						|
    else if (j->flags & HEIM_JSON_F_INDENT4)
 | 
						|
        while (i--)
 | 
						|
            j->out(j->ctx, "    ");
 | 
						|
    else if (j->flags & HEIM_JSON_F_INDENT8)
 | 
						|
        while (i--)
 | 
						|
            j->out(j->ctx, "        ");
 | 
						|
    else
 | 
						|
        while (i--)
 | 
						|
            j->out(j->ctx, "\t");
 | 
						|
}
 | 
						|
 | 
						|
static void
 | 
						|
array2json(heim_object_t value, void *ctx, int *stop)
 | 
						|
{
 | 
						|
    struct twojson *j = ctx;
 | 
						|
    if (j->ret)
 | 
						|
	return;
 | 
						|
    if (j->first) {
 | 
						|
	j->first = 0;
 | 
						|
    } else {
 | 
						|
	j->out(j->ctx, NULL); /* eat previous '\n' if possible */
 | 
						|
	j->out(j->ctx, ",\n");
 | 
						|
    }
 | 
						|
    j->ret = base2json(value, j, 0);
 | 
						|
}
 | 
						|
 | 
						|
static void
 | 
						|
dict2json(heim_object_t key, heim_object_t value, void *ctx)
 | 
						|
{
 | 
						|
    struct twojson *j = ctx;
 | 
						|
    if (j->ret)
 | 
						|
	return;
 | 
						|
    if (j->first) {
 | 
						|
	j->first = 0;
 | 
						|
    } else {
 | 
						|
	j->out(j->ctx, NULL); /* eat previous '\n' if possible */
 | 
						|
	j->out(j->ctx, ",\n");
 | 
						|
    }
 | 
						|
    j->ret = base2json(key, j, 0);
 | 
						|
    if (j->ret)
 | 
						|
	return;
 | 
						|
    switch (heim_get_tid(value)) {
 | 
						|
    case HEIM_TID_ARRAY:
 | 
						|
    case HEIM_TID_DICT:
 | 
						|
    case HEIM_TID_DATA:
 | 
						|
        j->out(j->ctx, ":\n");
 | 
						|
        j->indent++;
 | 
						|
        j->ret = base2json(value, j, 0);
 | 
						|
        if (j->ret)
 | 
						|
            return;
 | 
						|
        j->indent--;
 | 
						|
        break;
 | 
						|
    default:
 | 
						|
        j->out(j->ctx, ": ");
 | 
						|
        j->ret = base2json(value, j, 1);
 | 
						|
        break;
 | 
						|
    }
 | 
						|
}
 | 
						|
 | 
						|
#ifndef WIN32
 | 
						|
static void
 | 
						|
init_is_utf8(void *ptr)
 | 
						|
{
 | 
						|
    *(int *)ptr = strcasecmp("utf-8", nl_langinfo(CODESET)) == 0;
 | 
						|
}
 | 
						|
#endif
 | 
						|
 | 
						|
int
 | 
						|
heim_locale_is_utf8(void)
 | 
						|
{
 | 
						|
#ifdef WIN32
 | 
						|
    return 0; /* XXX Implement */
 | 
						|
#else
 | 
						|
    static int locale_is_utf8 = -1;
 | 
						|
    static heim_base_once_t once = HEIM_BASE_ONCE_INIT;
 | 
						|
 | 
						|
    heim_base_once_f(&once, &locale_is_utf8, init_is_utf8);
 | 
						|
    return locale_is_utf8;
 | 
						|
#endif
 | 
						|
}
 | 
						|
 | 
						|
static void
 | 
						|
out_escaped_bmp(struct twojson *j, const unsigned char *p, int nbytes)
 | 
						|
{
 | 
						|
    unsigned char e[sizeof("\\u0000")];
 | 
						|
    unsigned codepoint;
 | 
						|
 | 
						|
    if (nbytes == 2)
 | 
						|
        codepoint = ((p[0] & 0x1f) << 6) | (p[1] & 0x3f);
 | 
						|
    else if (nbytes == 3)
 | 
						|
        codepoint = ((p[0] & 0x0f) << 12) | ((p[1] & 0x3f) << 6) | (p[2] & 0x3f);
 | 
						|
    else
 | 
						|
        abort();
 | 
						|
    e[0]  = '\\';
 | 
						|
    e[1]  = 'u';
 | 
						|
    e[2]  = codepoint >> 12;
 | 
						|
    e[2] += (e[2] < 10) ? '0' : ('A' - 10);
 | 
						|
    e[3]  = (codepoint >> 8) & 0x0f;
 | 
						|
    e[3] += (e[3] < 10) ? '0' : ('A' - 10);
 | 
						|
    e[4]  = (codepoint >> 4) & 0x0f;
 | 
						|
    e[4] += (e[4] < 10) ? '0' : ('A' - 10);
 | 
						|
    e[5]  =  codepoint       & 0x0f;
 | 
						|
    e[5] += (e[5] < 10) ? '0' : ('A' - 10);
 | 
						|
    e[6]  = '\0';
 | 
						|
    j->out(j->ctx, (char *)e);
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
base2json(heim_object_t obj, struct twojson *j, int skip_indent)
 | 
						|
{
 | 
						|
    heim_tid_t type;
 | 
						|
    int first = 0;
 | 
						|
 | 
						|
    if (obj == NULL) {
 | 
						|
	if (j->flags & HEIM_JSON_F_CNULL2JSNULL) {
 | 
						|
	    obj = heim_null_create();
 | 
						|
	} else if (j->flags & HEIM_JSON_F_NO_C_NULL) {
 | 
						|
	    return EINVAL;
 | 
						|
	} else {
 | 
						|
	    indent(j);
 | 
						|
	    j->out(j->ctx, "<NULL>\n"); /* This is NOT valid JSON! */
 | 
						|
	    return 0;
 | 
						|
	}
 | 
						|
    }
 | 
						|
 | 
						|
    type = heim_get_tid(obj);
 | 
						|
    switch (type) {
 | 
						|
    case HEIM_TID_ARRAY:
 | 
						|
	indent(j);
 | 
						|
	j->out(j->ctx, "[\n");
 | 
						|
	j->indent++;
 | 
						|
	first = j->first;
 | 
						|
	j->first = 1;
 | 
						|
	heim_array_iterate_f(obj, j, array2json);
 | 
						|
	j->indent--;
 | 
						|
	if (!j->first)
 | 
						|
	    j->out(j->ctx, "\n");
 | 
						|
	indent(j);
 | 
						|
	j->out(j->ctx, "]\n");
 | 
						|
	j->first = first;
 | 
						|
	break;
 | 
						|
 | 
						|
    case HEIM_TID_DICT:
 | 
						|
	indent(j);
 | 
						|
	j->out(j->ctx, "{\n");
 | 
						|
	j->indent++;
 | 
						|
	first = j->first;
 | 
						|
	j->first = 1;
 | 
						|
	heim_dict_iterate_f(obj, j, dict2json);
 | 
						|
	j->indent--;
 | 
						|
	if (!j->first)
 | 
						|
	    j->out(j->ctx, "\n");
 | 
						|
	indent(j);
 | 
						|
	j->out(j->ctx, "}\n");
 | 
						|
	j->first = first;
 | 
						|
	break;
 | 
						|
 | 
						|
    case HEIM_TID_STRING: {
 | 
						|
	const unsigned char *s = (const unsigned char *)heim_string_get_utf8(obj);
 | 
						|
	const unsigned char *p;
 | 
						|
        unsigned int c, cp, ctop, cbot;
 | 
						|
        char e[sizeof("\\u0123\\u3210")];
 | 
						|
        int good;
 | 
						|
        size_t i;
 | 
						|
 | 
						|
        if (!skip_indent)
 | 
						|
            indent(j);
 | 
						|
	j->out(j->ctx, "\"");
 | 
						|
        for (p = s; (c = *p); p++) {
 | 
						|
            switch (c) {
 | 
						|
            /* ASCII control characters w/ C-like escapes */
 | 
						|
            case '\b': j->out(j->ctx, "\\b");  continue;
 | 
						|
            case '\f': j->out(j->ctx, "\\f");  continue;
 | 
						|
            case '\n': j->out(j->ctx, "\\n");  continue;
 | 
						|
            case '\r': j->out(j->ctx, "\\r");  continue;
 | 
						|
            case '\t': j->out(j->ctx, "\\t");  continue;
 | 
						|
            /* Other must-escape non-control ASCII characters */
 | 
						|
            case '"':  j->out(j->ctx, "\\\""); continue;
 | 
						|
            case '\\': j->out(j->ctx, "\\\\"); continue;
 | 
						|
            default: break;
 | 
						|
            }
 | 
						|
 | 
						|
            /*
 | 
						|
             * JSON string encoding is... complex.
 | 
						|
             *
 | 
						|
             * Invalid UTF-8 w/  HEIM_JSON_F_STRICT_STRINGS set -> return 1
 | 
						|
             *
 | 
						|
             * Invalid UTF-8 w/o HEIM_JSON_F_STRICT_STRINGS set -> pass
 | 
						|
             * through, a sort of Heimdal WTF-8, but not _the_ WTF-8.
 | 
						|
             */
 | 
						|
            if (c < 0x20) {
 | 
						|
                /* ASCII control character w/o C-like escape */
 | 
						|
                e[0] = '\\';
 | 
						|
                e[1] = 'u';
 | 
						|
                e[2] = '0';
 | 
						|
                e[3] = '0';
 | 
						|
                e[4] = "0123456789ABCDEF"[c>>4];
 | 
						|
                e[5] = "0123456789ABCDEF"[c & 0x0f];
 | 
						|
                e[6] = '\0';
 | 
						|
                j->out(j->ctx, e);
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
            if (c < 0x80) {
 | 
						|
                /* ASCII */
 | 
						|
                e[0] = c;
 | 
						|
                e[1] = '\0';
 | 
						|
                j->out(j->ctx, e);
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
            if ((c & 0xc0) == 0x80) {
 | 
						|
                /* UTF-8 bare non-leading byte */
 | 
						|
                if (!(j->flags & HEIM_JSON_F_STRICT_STRINGS)) {
 | 
						|
                    e[0] = c;
 | 
						|
                    e[1] = '\0';
 | 
						|
                    j->out(j->ctx, e);
 | 
						|
                    continue;
 | 
						|
                }
 | 
						|
                return 1;
 | 
						|
            }
 | 
						|
            if ((c & 0xe0) == 0xc0) {
 | 
						|
                /* UTF-8 leading byte of two-byte sequence */
 | 
						|
                good = 1;
 | 
						|
                for (i = 1; i < 2 && good && p[i]; i++) {
 | 
						|
                    if ((p[i] & 0xc0) != 0x80)
 | 
						|
                        good = 0;
 | 
						|
                }
 | 
						|
                if (i != 2)
 | 
						|
                    good = 0;
 | 
						|
                if (!good && !(j->flags & HEIM_JSON_F_STRICT_STRINGS)) {
 | 
						|
                    e[0] = c;
 | 
						|
                    e[1] = '\0';
 | 
						|
                    j->out(j->ctx, e);
 | 
						|
                    continue;
 | 
						|
                } else if (!good) {
 | 
						|
                    return 1;
 | 
						|
                }
 | 
						|
                if (j->flags & HEIM_JSON_F_ESCAPE_NON_ASCII) {
 | 
						|
                    out_escaped_bmp(j, p, 2);
 | 
						|
                    p += 1;
 | 
						|
                    continue;
 | 
						|
                }
 | 
						|
                e[0] = c;
 | 
						|
                e[1] = p[1];
 | 
						|
                e[2] = '\0';
 | 
						|
                j->out(j->ctx, e);
 | 
						|
                p += 1;
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
            if ((c & 0xf0) == 0xe0) {
 | 
						|
                /* UTF-8 leading byte of three-byte sequence */
 | 
						|
                good = 1;
 | 
						|
                for (i = 1; i < 3 && good && p[i]; i++) {
 | 
						|
                    if ((p[i] & 0xc0) != 0x80)
 | 
						|
                        good = 0;
 | 
						|
                }
 | 
						|
                if (i != 3)
 | 
						|
                    good = 0;
 | 
						|
                if (!good && !(j->flags & HEIM_JSON_F_STRICT_STRINGS)) {
 | 
						|
                    e[0] = c;
 | 
						|
                    e[1] = '\0';
 | 
						|
                    j->out(j->ctx, e);
 | 
						|
                    continue;
 | 
						|
                } else if (!good) {
 | 
						|
                    return 1;
 | 
						|
                }
 | 
						|
                if (j->flags & HEIM_JSON_F_ESCAPE_NON_ASCII) {
 | 
						|
                    out_escaped_bmp(j, p, 3);
 | 
						|
                    p += 2;
 | 
						|
                    continue;
 | 
						|
                }
 | 
						|
                e[0] = c;
 | 
						|
                e[1] = p[1];
 | 
						|
                e[2] = p[2];
 | 
						|
                e[3] = '\0';
 | 
						|
                j->out(j->ctx, e);
 | 
						|
                p += 2;
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
 | 
						|
            if (c > 0xf7) {
 | 
						|
                /* Invalid UTF-8 leading byte */
 | 
						|
                if (!(j->flags & HEIM_JSON_F_STRICT_STRINGS)) {
 | 
						|
                    e[0] = c;
 | 
						|
                    e[1] = '\0';
 | 
						|
                    j->out(j->ctx, e);
 | 
						|
                    continue;
 | 
						|
                }
 | 
						|
                return 1;
 | 
						|
            }
 | 
						|
 | 
						|
            /*
 | 
						|
             * A codepoint > U+FFFF, needs encoding a la UTF-16 surrogate
 | 
						|
             * pair because JSON takes after JS which uses UTF-16.  Ugly.
 | 
						|
             */
 | 
						|
            cp = c & 0x7;
 | 
						|
            good = 1;
 | 
						|
            for (i = 1; i < 4 && good && p[i]; i++) {
 | 
						|
                if ((p[i] & 0xc0) == 0x80)
 | 
						|
                    cp = (cp << 6) | (p[i] & 0x3f);
 | 
						|
                else
 | 
						|
                    good = 0;
 | 
						|
            }
 | 
						|
            if (i != 4)
 | 
						|
                good = 0;
 | 
						|
            if (!good && !(j->flags & HEIM_JSON_F_STRICT_STRINGS)) {
 | 
						|
                e[0] = c;
 | 
						|
                e[1] = '\0';
 | 
						|
                j->out(j->ctx, e);
 | 
						|
                continue;
 | 
						|
            } else if (!good) {
 | 
						|
                return 1;
 | 
						|
            }
 | 
						|
            p += 3;
 | 
						|
 | 
						|
            cp -= 0x10000;
 | 
						|
            ctop = 0xD800 + (cp >>   10);
 | 
						|
            cbot = 0xDC00 + (cp & 0x3ff);
 | 
						|
 | 
						|
            e[0 ] = '\\';
 | 
						|
            e[1 ] = 'u';
 | 
						|
            e[2 ] = "0123456789ABCDEF"[(ctop         ) >> 12];
 | 
						|
            e[3 ] = "0123456789ABCDEF"[(ctop & 0x0f00) >>  8];
 | 
						|
            e[4 ] = "0123456789ABCDEF"[(ctop & 0x00f0) >>  4];
 | 
						|
            e[5 ] = "0123456789ABCDEF"[(ctop & 0x000f)      ];
 | 
						|
            e[6 ] = '\\';
 | 
						|
            e[7 ] = 'u';
 | 
						|
            e[8 ] = "0123456789ABCDEF"[(cbot         ) >> 12];
 | 
						|
            e[9 ] = "0123456789ABCDEF"[(cbot & 0x0f00) >>  8];
 | 
						|
            e[10] = "0123456789ABCDEF"[(cbot & 0x00f0) >>  4];
 | 
						|
            e[11] = "0123456789ABCDEF"[(cbot & 0x000f)      ];
 | 
						|
            e[12] = '\0';
 | 
						|
            j->out(j->ctx, e);
 | 
						|
            continue;
 | 
						|
        }
 | 
						|
	j->out(j->ctx, "\"");
 | 
						|
	break;
 | 
						|
    }
 | 
						|
 | 
						|
    case HEIM_TID_DATA: {
 | 
						|
	heim_dict_t d;
 | 
						|
	heim_string_t v;
 | 
						|
	const heim_octet_string *data;
 | 
						|
	char *b64 = NULL;
 | 
						|
	int ret;
 | 
						|
 | 
						|
	if (j->flags & HEIM_JSON_F_NO_DATA)
 | 
						|
	    return EINVAL; /* JSON doesn't do binary */
 | 
						|
 | 
						|
	data = heim_data_get_data(obj);
 | 
						|
	ret = rk_base64_encode(data->data, data->length, &b64);
 | 
						|
	if (ret < 0 || b64 == NULL)
 | 
						|
	    return ENOMEM;
 | 
						|
 | 
						|
	if (j->flags & HEIM_JSON_F_NO_DATA_DICT) {
 | 
						|
	    indent(j);
 | 
						|
	    j->out(j->ctx, "\"");
 | 
						|
	    j->out(j->ctx, b64); /* base64-encode; hope there's no aliasing */
 | 
						|
	    j->out(j->ctx, "\"");
 | 
						|
	    free(b64);
 | 
						|
	} else {
 | 
						|
	    /*
 | 
						|
	     * JSON has no way to represent binary data, therefore the
 | 
						|
	     * following is a Heimdal-specific convention.
 | 
						|
	     *
 | 
						|
	     * We encode binary data as a dict with a single very magic
 | 
						|
	     * key with a base64-encoded value.  The magic key includes
 | 
						|
	     * a uuid, so we're not likely to alias accidentally.
 | 
						|
	     */
 | 
						|
	    d = heim_dict_create(2);
 | 
						|
	    if (d == NULL) {
 | 
						|
		free(b64);
 | 
						|
		return ENOMEM;
 | 
						|
	    }
 | 
						|
	    v = heim_string_ref_create(b64, free);
 | 
						|
	    if (v == NULL) {
 | 
						|
		free(b64);
 | 
						|
		heim_release(d);
 | 
						|
		return ENOMEM;
 | 
						|
	    }
 | 
						|
	    ret = heim_dict_set_value(d, heim_tid_data_uuid_key, v);
 | 
						|
	    heim_release(v);
 | 
						|
	    if (ret) {
 | 
						|
		heim_release(d);
 | 
						|
		return ENOMEM;
 | 
						|
	    }
 | 
						|
	    ret = base2json(d, j, 0);
 | 
						|
	    heim_release(d);
 | 
						|
	    if (ret)
 | 
						|
		return ret;
 | 
						|
	}
 | 
						|
	break;
 | 
						|
    }
 | 
						|
 | 
						|
    case HEIM_TID_NUMBER: {
 | 
						|
	char num[32];
 | 
						|
        if (!skip_indent)
 | 
						|
            indent(j);
 | 
						|
	snprintf(num, sizeof (num), "%d", heim_number_get_int(obj));
 | 
						|
	j->out(j->ctx, num);
 | 
						|
	break;
 | 
						|
    }
 | 
						|
    case HEIM_TID_NULL:
 | 
						|
        if (!skip_indent)
 | 
						|
            indent(j);
 | 
						|
	j->out(j->ctx, "null");
 | 
						|
	break;
 | 
						|
    case HEIM_TID_BOOL:
 | 
						|
        if (!skip_indent)
 | 
						|
            indent(j);
 | 
						|
	j->out(j->ctx, heim_bool_val(obj) ? "true" : "false");
 | 
						|
	break;
 | 
						|
    default:
 | 
						|
	return 1;
 | 
						|
    }
 | 
						|
    return 0;
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
heim_base2json(heim_object_t obj, void *ctx, heim_json_flags_t flags,
 | 
						|
	       void (*out)(void *, const char *))
 | 
						|
{
 | 
						|
    struct twojson j;
 | 
						|
 | 
						|
    heim_base_once_f(&heim_json_once, NULL, json_init_once);
 | 
						|
 | 
						|
    j.indent = 0;
 | 
						|
    j.ctx = ctx;
 | 
						|
    j.out = out;
 | 
						|
    j.flags = flags;
 | 
						|
    j.ret = 0;
 | 
						|
    j.first = 1;
 | 
						|
 | 
						|
    if (!(flags & HEIM_JSON_F_NO_ESCAPE_NON_ASCII) &&
 | 
						|
        !heim_locale_is_utf8())
 | 
						|
        j.flags |= HEIM_JSON_F_ESCAPE_NON_ASCII;
 | 
						|
 | 
						|
    return base2json(obj, &j, 0);
 | 
						|
}
 | 
						|
 | 
						|
 | 
						|
/*
 | 
						|
 *
 | 
						|
 */
 | 
						|
 | 
						|
struct parse_ctx {
 | 
						|
    unsigned long lineno;
 | 
						|
    const uint8_t *p;
 | 
						|
    const uint8_t *pstart;
 | 
						|
    const uint8_t *pend;
 | 
						|
    heim_error_t error;
 | 
						|
    size_t depth;
 | 
						|
    heim_json_flags_t flags;
 | 
						|
};
 | 
						|
 | 
						|
 | 
						|
static heim_object_t
 | 
						|
parse_value(struct parse_ctx *ctx);
 | 
						|
 | 
						|
/*
 | 
						|
 * This function eats whitespace, but, critically, it also succeeds
 | 
						|
 * only if there's anything left to parse.
 | 
						|
 */
 | 
						|
static int
 | 
						|
white_spaces(struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    while (ctx->p < ctx->pend) {
 | 
						|
	uint8_t c = *ctx->p;
 | 
						|
	if (c == ' ' || c == '\t' || c == '\r') {
 | 
						|
 | 
						|
	} else if (c == '\n') {
 | 
						|
	    ctx->lineno++;
 | 
						|
	} else
 | 
						|
	    return 0;
 | 
						|
	(ctx->p)++;
 | 
						|
    }
 | 
						|
    return -1;
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
is_number(uint8_t n)
 | 
						|
{
 | 
						|
    return ('0' <= n && n <= '9');
 | 
						|
}
 | 
						|
 | 
						|
static heim_number_t
 | 
						|
parse_number(struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    int number = 0, neg = 1;
 | 
						|
 | 
						|
    if (ctx->p >= ctx->pend)
 | 
						|
	return NULL;
 | 
						|
 | 
						|
    if (*ctx->p == '-') {
 | 
						|
	if (ctx->p + 1 >= ctx->pend)
 | 
						|
	    return NULL;
 | 
						|
	neg = -1;
 | 
						|
	ctx->p += 1;
 | 
						|
    }
 | 
						|
 | 
						|
    while (ctx->p < ctx->pend) {
 | 
						|
	if (is_number(*ctx->p)) {
 | 
						|
	    number = (number * 10) + (*ctx->p - '0');
 | 
						|
	} else {
 | 
						|
	    break;
 | 
						|
	}
 | 
						|
	ctx->p += 1;
 | 
						|
    }
 | 
						|
 | 
						|
    return heim_number_create(number * neg);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Read 4 hex digits from ctx->p.
 | 
						|
 *
 | 
						|
 * If we don't have enough, rewind ctx->p and return -1 .
 | 
						|
 */
 | 
						|
static int
 | 
						|
unescape_unicode(struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    int c = 0;
 | 
						|
    int i;
 | 
						|
 | 
						|
    for (i = 0; i < 4 && ctx->p < ctx->pend; i++, ctx->p++) {
 | 
						|
        if (*ctx->p >= '0' && *ctx->p <= '9') {
 | 
						|
            c = (c << 4) + (*ctx->p - '0');
 | 
						|
        } else if (*ctx->p >= 'A' && *ctx->p <= 'F') {
 | 
						|
            c = (c << 4) + (10 + *ctx->p - 'A');
 | 
						|
        } else if (*ctx->p >= 'a' && *ctx->p <= 'f') {
 | 
						|
            c = (c << 4) + (10 + *ctx->p - 'a');
 | 
						|
        } else {
 | 
						|
            ctx->p -= i;
 | 
						|
            return -1;
 | 
						|
        }
 | 
						|
    }
 | 
						|
    return c;
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
encode_utf8(struct parse_ctx *ctx, char **pp, char *pend, int c)
 | 
						|
{
 | 
						|
    char *p = *pp;
 | 
						|
 | 
						|
    if (c < 0x80) {
 | 
						|
        /* ASCII */
 | 
						|
        if (p >= pend) return 0;
 | 
						|
        *(p++) = c;
 | 
						|
        *pp = p;
 | 
						|
        return 1;
 | 
						|
    }
 | 
						|
    if (c < 0x800) {
 | 
						|
        /* 2 code unit UTF-8 sequence */
 | 
						|
        if (p >= pend) return 0;
 | 
						|
        *(p++) = 0xc0 | ((c >>  6)       );
 | 
						|
        if (p == pend) return 0;
 | 
						|
        *(p++) = 0x80 | ((c      ) & 0x3f);
 | 
						|
        *pp = p;
 | 
						|
        return 1;
 | 
						|
    }
 | 
						|
    if (c < 0x10000) {
 | 
						|
        /* 3 code unit UTF-8 sequence */
 | 
						|
        if (p >= pend) return 0;
 | 
						|
        *(p++) = 0xe0 | ((c >> 12)       );
 | 
						|
        if (p == pend) return 0;
 | 
						|
        *(p++) = 0x80 | ((c >>  6) & 0x3f);
 | 
						|
        if (p == pend) return 0;
 | 
						|
        *(p++) = 0x80 | ((c)       & 0x3f);
 | 
						|
        *pp = p;
 | 
						|
        return 1;
 | 
						|
    }
 | 
						|
    if (c < 0x110000) {
 | 
						|
        /* 4 code unit UTF-8 sequence */
 | 
						|
        if (p >= pend) return 0;
 | 
						|
        *(p++) = 0xf0 | ((c >> 18)       );
 | 
						|
        if (p == pend) return 0;
 | 
						|
        *(p++) = 0x80 | ((c >> 12) & 0x3f);
 | 
						|
        if (p == pend) return 0;
 | 
						|
        *(p++) = 0x80 | ((c >>  6) & 0x3f);
 | 
						|
        if (p == pend) return 0;
 | 
						|
        *(p++) = 0x80 | ((c)       & 0x3f);
 | 
						|
        *pp = p;
 | 
						|
        return 1;
 | 
						|
    }
 | 
						|
    return 0;
 | 
						|
}
 | 
						|
 | 
						|
static heim_string_t
 | 
						|
parse_string_error(struct parse_ctx *ctx,
 | 
						|
                   char *freeme,
 | 
						|
                   const char *msg)
 | 
						|
{
 | 
						|
    free(freeme);
 | 
						|
    ctx->error = heim_error_create(EINVAL, "%s at %lu", msg, ctx->lineno);
 | 
						|
    return NULL;
 | 
						|
}
 | 
						|
 | 
						|
static heim_string_t
 | 
						|
parse_string(struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    const uint8_t *start;
 | 
						|
    heim_object_t o;
 | 
						|
    size_t alloc_len = 0;
 | 
						|
    size_t need = 0;
 | 
						|
    char *p0, *p, *pend;
 | 
						|
    int strict = ctx->flags & HEIM_JSON_F_STRICT_STRINGS;
 | 
						|
    int binary = 0;
 | 
						|
 | 
						|
    if (*ctx->p != '"')
 | 
						|
        return parse_string_error(ctx, NULL,
 | 
						|
                                  "Expected a JSON string but found "
 | 
						|
                                  "something else");
 | 
						|
    start = ++(ctx->p);
 | 
						|
 | 
						|
    /* Estimate how many bytes we need to allocate */
 | 
						|
    p0 = p = pend = NULL;
 | 
						|
    for (need = 1; ctx->p < ctx->pend; ctx->p++) {
 | 
						|
        need++;
 | 
						|
        if (*ctx->p == '\\')
 | 
						|
            ctx->p++;
 | 
						|
        else if (*ctx->p == '"')
 | 
						|
            break;
 | 
						|
    }
 | 
						|
    if (ctx->p == ctx->pend)
 | 
						|
        return parse_string_error(ctx, NULL, "Unterminated JSON string");
 | 
						|
 | 
						|
    ctx->p = start;
 | 
						|
    while (ctx->p < ctx->pend) {
 | 
						|
        const unsigned char *p_save;
 | 
						|
        int32_t ctop, cbot;
 | 
						|
 | 
						|
        if (*ctx->p == '"') {
 | 
						|
            ctx->p++;
 | 
						|
            break;
 | 
						|
        }
 | 
						|
 | 
						|
        /* Allocate or resize our output buffer if need be */
 | 
						|
        if (need || p == pend) {
 | 
						|
            char *tmp;
 | 
						|
 | 
						|
            /*
 | 
						|
             * Work out how far p is into p0 to re-esablish p after
 | 
						|
             * the realloc()
 | 
						|
             */
 | 
						|
            size_t p0_to_p_len = (p - p0);
 | 
						|
 | 
						|
            tmp = realloc(p0, alloc_len + need + 5 /* slop? */);
 | 
						|
 | 
						|
            if (tmp == NULL) {
 | 
						|
                ctx->error = heim_error_create_enomem();
 | 
						|
                free(p0);
 | 
						|
                return NULL;
 | 
						|
            }
 | 
						|
            alloc_len += need + 5;
 | 
						|
 | 
						|
            /*
 | 
						|
             * We have two pointers, p and p0, we want to keep them
 | 
						|
             * pointing into the same memory after the realloc()
 | 
						|
             */
 | 
						|
            p = tmp + p0_to_p_len;
 | 
						|
            p0 = tmp;
 | 
						|
            pend = p0 + alloc_len;
 | 
						|
 | 
						|
            need = 0;
 | 
						|
        }
 | 
						|
 | 
						|
        if (*ctx->p != '\\') {
 | 
						|
            unsigned char c = *ctx->p;
 | 
						|
 | 
						|
            /*
 | 
						|
             * Not backslashed -> consume now.
 | 
						|
             *
 | 
						|
             * NOTE: All cases in this block must continue or return w/ error.
 | 
						|
             */
 | 
						|
 | 
						|
            /* Check for unescaped ASCII control characters */
 | 
						|
            if (c == '\n') {
 | 
						|
                if (strict)
 | 
						|
                    return parse_string_error(ctx, p0,
 | 
						|
                                              "Unescaped newline in JSON string");
 | 
						|
                /* Count the newline but don't add it to the decoding */
 | 
						|
                ctx->lineno++;
 | 
						|
            } else if (strict && *ctx->p <= 0x1f) {
 | 
						|
                return parse_string_error(ctx, p0, "Unescaped ASCII control character");
 | 
						|
            } else if (c == 0) {
 | 
						|
                binary = 1;
 | 
						|
            }
 | 
						|
            if (!strict || c < 0x80) {
 | 
						|
                /* ASCII, or not strict -> no need to validate */
 | 
						|
                *(p++) = c;
 | 
						|
                ctx->p++;
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
 | 
						|
            /*
 | 
						|
             * Being strict for parsing means we want to detect malformed UTF-8
 | 
						|
             * sequences.
 | 
						|
             *
 | 
						|
             * If not strict then we just go on below and add to `p' whatever
 | 
						|
             * bytes we find in `ctx->p' as we find them.
 | 
						|
             *
 | 
						|
             * For each two-byte sequence we need one more byte in `p[]'.  For
 | 
						|
             * each three-byte sequence we need two more bytes in `p[]'.
 | 
						|
             *
 | 
						|
             * Setting `need' and looping will cause `p0' to be grown.
 | 
						|
             *
 | 
						|
             * NOTE: All cases in this block must continue or return w/ error.
 | 
						|
             */
 | 
						|
            if ((c & 0xe0) == 0xc0) {
 | 
						|
                /* Two-byte UTF-8 encoding */
 | 
						|
                if (pend - p < 2) {
 | 
						|
                    need = 2;
 | 
						|
                    continue; /* realloc p0 */
 | 
						|
                }
 | 
						|
 | 
						|
                *(p++) = c;
 | 
						|
                ctx->p++;
 | 
						|
                if (ctx->p == ctx->pend)
 | 
						|
                    return parse_string_error(ctx, p0, "Truncated UTF-8");
 | 
						|
                c = *(ctx->p++);
 | 
						|
                if ((c & 0xc0) != 0x80)
 | 
						|
                    return parse_string_error(ctx, p0, "Truncated UTF-8");
 | 
						|
                *(p++) = c;
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
            if ((c & 0xf0) == 0xe0) {
 | 
						|
                /* Three-byte UTF-8 encoding */
 | 
						|
                if (pend - p < 3) {
 | 
						|
                    need = 3;
 | 
						|
                    continue; /* realloc p0 */
 | 
						|
                }
 | 
						|
 | 
						|
                *(p++) = c;
 | 
						|
                ctx->p++;
 | 
						|
                if (ctx->p == ctx->pend)
 | 
						|
                    return parse_string_error(ctx, p0, "Truncated UTF-8");
 | 
						|
                c = *(ctx->p++);
 | 
						|
                if ((c & 0xc0) != 0x80)
 | 
						|
                    return parse_string_error(ctx, p0, "Truncated UTF-8");
 | 
						|
                *(p++) = c;
 | 
						|
                c = *(ctx->p++);
 | 
						|
                if ((c & 0xc0) != 0x80)
 | 
						|
                    return parse_string_error(ctx, p0, "Truncated UTF-8");
 | 
						|
                *(p++) = c;
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
            if ((c & 0xf8) == 0xf0)
 | 
						|
                return parse_string_error(ctx, p0, "UTF-8 sequence not "
 | 
						|
                                          "encoded as escaped UTF-16");
 | 
						|
            if ((c & 0xc0) == 0x80)
 | 
						|
                return parse_string_error(ctx, p0,
 | 
						|
                                          "Invalid UTF-8 "
 | 
						|
                                          "(bare continuation code unit)");
 | 
						|
 | 
						|
            return parse_string_error(ctx, p0, "Not UTF-8");
 | 
						|
        }
 | 
						|
 | 
						|
        /* Backslash-quoted character */
 | 
						|
        ctx->p++;
 | 
						|
        if (ctx->p == ctx->pend) {
 | 
						|
            ctx->error =
 | 
						|
                heim_error_create(EINVAL,
 | 
						|
                                  "Unterminated JSON string at line %lu",
 | 
						|
                                  ctx->lineno);
 | 
						|
            free(p0);
 | 
						|
            return NULL;
 | 
						|
        }
 | 
						|
        switch (*ctx->p) {
 | 
						|
        /* Simple escapes */
 | 
						|
        case  'b': *(p++) = '\b'; ctx->p++; continue;
 | 
						|
        case  'f': *(p++) = '\f'; ctx->p++; continue;
 | 
						|
        case  'n': *(p++) = '\n'; ctx->p++; continue;
 | 
						|
        case  'r': *(p++) = '\r'; ctx->p++; continue;
 | 
						|
        case  't': *(p++) = '\t'; ctx->p++; continue;
 | 
						|
        case  '"': *(p++) = '"';  ctx->p++; continue;
 | 
						|
        case '\\': *(p++) = '\\'; ctx->p++; continue;
 | 
						|
        /* Escaped Unicode handled below */
 | 
						|
        case  'u':
 | 
						|
            /*
 | 
						|
             * Worst case for !strict we need 11 bytes for a truncated non-BMP
 | 
						|
             * codepoint escape.  Call it 12.
 | 
						|
             */
 | 
						|
            if (strict)
 | 
						|
                need = 4;
 | 
						|
            else
 | 
						|
                need = 12;
 | 
						|
            if (pend - p < need) {
 | 
						|
                /* Go back to the backslash, realloc, try again */
 | 
						|
                ctx->p--;
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
 | 
						|
            need = 0;
 | 
						|
            ctx->p++;
 | 
						|
            break;
 | 
						|
        default:
 | 
						|
            if (!strict) {
 | 
						|
                *(p++) = *ctx->p;
 | 
						|
                ctx->p++;
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
            ctx->error =
 | 
						|
                heim_error_create(EINVAL,
 | 
						|
                                  "Invalid backslash escape at line %lu",
 | 
						|
                                  ctx->lineno);
 | 
						|
            free(p0);
 | 
						|
            return NULL;
 | 
						|
        }
 | 
						|
 | 
						|
        /* Unicode code point */
 | 
						|
        if (pend - p < 12) {
 | 
						|
            need = 12;
 | 
						|
            ctx->p -= 2; /* for "\\u" */
 | 
						|
            continue; /* This will cause p0 to be realloc'ed */
 | 
						|
        }
 | 
						|
        p_save = ctx->p;
 | 
						|
        cbot = -3;
 | 
						|
        ctop = unescape_unicode(ctx);
 | 
						|
        if (ctop == -1 && strict)
 | 
						|
            return parse_string_error(ctx, p0, "Invalid escaped Unicode");
 | 
						|
        if (ctop == -1) {
 | 
						|
            /*
 | 
						|
             * Not strict; tolerate bad input.
 | 
						|
             *
 | 
						|
             * Output "\\u" and then loop to treat what we expected to be four
 | 
						|
             * digits as if they were not part of an escaped Unicode codepoint.
 | 
						|
             */
 | 
						|
            ctx->p = p_save;
 | 
						|
            if (p < pend)
 | 
						|
                *(p++) = '\\';
 | 
						|
            if (p < pend)
 | 
						|
                *(p++) = 'u';
 | 
						|
            continue;
 | 
						|
        }
 | 
						|
        if (ctop == 0) {
 | 
						|
            *(p++) = '\0';
 | 
						|
            binary = 1;
 | 
						|
            continue;
 | 
						|
        }
 | 
						|
        if (ctop < 0xd800) {
 | 
						|
            if (!encode_utf8(ctx, &p, pend, ctop))
 | 
						|
                return parse_string_error(ctx, p0,
 | 
						|
                                          "Internal JSON string parse error");
 | 
						|
            continue;
 | 
						|
        }
 | 
						|
 | 
						|
        /*
 | 
						|
         * We parsed the top escaped codepoint of a surrogate pair encoding
 | 
						|
         * of a non-BMP Unicode codepoint.  What follows must be another
 | 
						|
         * escaped codepoint.
 | 
						|
         */
 | 
						|
        if (ctx->p < ctx->pend && ctx->p[0] == '\\')
 | 
						|
            ctx->p++;
 | 
						|
        else
 | 
						|
            ctop = -1;
 | 
						|
        if (ctop > -1 && ctx->p < ctx->pend && ctx->p[0] == 'u')
 | 
						|
            ctx->p++;
 | 
						|
        else
 | 
						|
            ctop = -1;
 | 
						|
        if (ctop > -1) {
 | 
						|
            /* Parse the hex digits of the bottom half of the surrogate pair */
 | 
						|
            cbot = unescape_unicode(ctx);
 | 
						|
            if (cbot == -1 || cbot < 0xdc00)
 | 
						|
                ctop = -1;
 | 
						|
        }
 | 
						|
        if (ctop == -1) {
 | 
						|
            if (strict)
 | 
						|
                return parse_string_error(ctx, p0,
 | 
						|
                                          "Invalid surrogate pair");
 | 
						|
 | 
						|
            /*
 | 
						|
             * Output "\\u", rewind, output the digits of `ctop'.
 | 
						|
             *
 | 
						|
             * When we get to what should have been the bottom half of the
 | 
						|
             * pair we'll necessarily fail to parse it as a normal escaped
 | 
						|
             * Unicode codepoint, and once again, rewind and output its digits.
 | 
						|
             */
 | 
						|
            if (p < pend)
 | 
						|
                *(p++) = '\\';
 | 
						|
            if (p < pend)
 | 
						|
                *(p++) = 'u';
 | 
						|
            ctx->p = p_save;
 | 
						|
            continue;
 | 
						|
        }
 | 
						|
 | 
						|
        /* Finally decode the surrogate pair then encode as UTF-8 */
 | 
						|
        ctop -= 0xd800;
 | 
						|
        cbot -= 0xdc00;
 | 
						|
        if (!encode_utf8(ctx, &p, pend, 0x10000 + ((ctop << 10) | (cbot & 0x3ff))))
 | 
						|
            return parse_string_error(ctx, p0,
 | 
						|
                                      "Internal JSON string parse error");
 | 
						|
    }
 | 
						|
 | 
						|
    if (p0 == NULL)
 | 
						|
        return heim_string_create("");
 | 
						|
 | 
						|
    /* NUL-terminate for rk_base64_decode() and plain paranoia */
 | 
						|
    if (p0 != NULL && p == pend) {
 | 
						|
        /*
 | 
						|
	 * Work out how far p is into p0 to re-esablish p after
 | 
						|
	 * the realloc()
 | 
						|
	 */
 | 
						|
        size_t p0_to_pend_len = (pend - p0);
 | 
						|
        char *tmp = realloc(p0, 1 + p0_to_pend_len);
 | 
						|
 | 
						|
        if (tmp == NULL) {
 | 
						|
            ctx->error = heim_error_create_enomem();
 | 
						|
            free(p0);
 | 
						|
            return NULL;
 | 
						|
        }
 | 
						|
        /*
 | 
						|
         * We have three pointers, p, pend (which are the same)
 | 
						|
         * and p0, we want to keep them pointing into the same
 | 
						|
         * memory after the realloc()
 | 
						|
         */
 | 
						|
        p = tmp + p0_to_pend_len;
 | 
						|
 | 
						|
        pend = p + 1;
 | 
						|
        p0 = tmp;
 | 
						|
    }
 | 
						|
    *(p++) = '\0';
 | 
						|
 | 
						|
    /* If there's embedded NULs, it's not a C string */
 | 
						|
    if (binary) {
 | 
						|
        o = heim_data_ref_create(p0, (p - 1) - p0, free);
 | 
						|
        return o;
 | 
						|
    }
 | 
						|
 | 
						|
    /* Sadly this will copy `p0' */
 | 
						|
    o = heim_string_create_with_bytes(p0, p - p0);
 | 
						|
    free(p0);
 | 
						|
    return o;
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
parse_pair(heim_dict_t dict, struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    heim_string_t key;
 | 
						|
    heim_object_t value;
 | 
						|
 | 
						|
    if (white_spaces(ctx))
 | 
						|
	return -1;
 | 
						|
 | 
						|
    if (*ctx->p == '}') {
 | 
						|
	ctx->p++;
 | 
						|
	return 0;
 | 
						|
    }
 | 
						|
 | 
						|
    if (ctx->flags & HEIM_JSON_F_STRICT_DICT)
 | 
						|
	/* JSON allows only string keys */
 | 
						|
	key = parse_string(ctx);
 | 
						|
    else
 | 
						|
	/* heim_dict_t allows any heim_object_t as key */
 | 
						|
	key = parse_value(ctx);
 | 
						|
    if (key == NULL)
 | 
						|
	/* Even heim_dict_t does not allow C NULLs as keys though! */
 | 
						|
	return -1;
 | 
						|
 | 
						|
    if (white_spaces(ctx)) {
 | 
						|
	heim_release(key);
 | 
						|
	return -1;
 | 
						|
    }
 | 
						|
 | 
						|
    if (*ctx->p != ':') {
 | 
						|
	heim_release(key);
 | 
						|
	return -1;
 | 
						|
    }
 | 
						|
 | 
						|
    ctx->p += 1; /* safe because we call white_spaces() next */
 | 
						|
 | 
						|
    if (white_spaces(ctx)) {
 | 
						|
	heim_release(key);
 | 
						|
	return -1;
 | 
						|
    }
 | 
						|
 | 
						|
    value = parse_value(ctx);
 | 
						|
    if (value == NULL &&
 | 
						|
	(ctx->error != NULL || (ctx->flags & HEIM_JSON_F_NO_C_NULL))) {
 | 
						|
	if (ctx->error == NULL)
 | 
						|
	    ctx->error = heim_error_create(EINVAL, "Invalid JSON encoding");
 | 
						|
	heim_release(key);
 | 
						|
	return -1;
 | 
						|
    }
 | 
						|
    heim_dict_set_value(dict, key, value);
 | 
						|
    heim_release(key);
 | 
						|
    heim_release(value);
 | 
						|
 | 
						|
    if (white_spaces(ctx))
 | 
						|
	return -1;
 | 
						|
 | 
						|
    if (*ctx->p == '}') {
 | 
						|
	/*
 | 
						|
	 * Return 1 but don't consume the '}' so we can count the one
 | 
						|
	 * pair in a one-pair dict
 | 
						|
	 */
 | 
						|
	return 1;
 | 
						|
    } else if (*ctx->p == ',') {
 | 
						|
	ctx->p++;
 | 
						|
	return 1;
 | 
						|
    }
 | 
						|
    return -1;
 | 
						|
}
 | 
						|
 | 
						|
static heim_dict_t
 | 
						|
parse_dict(struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    heim_dict_t dict;
 | 
						|
    size_t count = 0;
 | 
						|
    int ret;
 | 
						|
 | 
						|
    heim_assert(*ctx->p == '{', "string doesn't start with {");
 | 
						|
 | 
						|
    dict = heim_dict_create(11);
 | 
						|
    if (dict == NULL) {
 | 
						|
	ctx->error = heim_error_create_enomem();
 | 
						|
	return NULL;
 | 
						|
    }
 | 
						|
 | 
						|
    ctx->p += 1; /* safe because parse_pair() calls white_spaces() first */
 | 
						|
 | 
						|
    while ((ret = parse_pair(dict, ctx)) > 0)
 | 
						|
	count++;
 | 
						|
    if (ret < 0) {
 | 
						|
	heim_release(dict);
 | 
						|
	return NULL;
 | 
						|
    }
 | 
						|
    if (count == 1 && !(ctx->flags & HEIM_JSON_F_NO_DATA_DICT)) {
 | 
						|
	heim_object_t v = heim_dict_copy_value(dict, heim_tid_data_uuid_key);
 | 
						|
 | 
						|
	/*
 | 
						|
	 * Binary data encoded as a dict with a single magic key with
 | 
						|
	 * base64-encoded value?  Decode as heim_data_t.
 | 
						|
	 */
 | 
						|
	if (v != NULL && heim_get_tid(v) == HEIM_TID_STRING) {
 | 
						|
	    void *buf;
 | 
						|
	    size_t len;
 | 
						|
 | 
						|
	    buf = malloc(strlen(heim_string_get_utf8(v)));
 | 
						|
	    if (buf == NULL) {
 | 
						|
		heim_release(dict);
 | 
						|
		heim_release(v);
 | 
						|
		ctx->error = heim_error_create_enomem();
 | 
						|
		return NULL;
 | 
						|
	    }
 | 
						|
	    len = rk_base64_decode(heim_string_get_utf8(v), buf);
 | 
						|
	    heim_release(v);
 | 
						|
	    if (len == -1) {
 | 
						|
		free(buf);
 | 
						|
		return dict; /* assume aliasing accident */
 | 
						|
	    }
 | 
						|
	    heim_release(dict);
 | 
						|
	    return (heim_dict_t)heim_data_ref_create(buf, len, free);
 | 
						|
	}
 | 
						|
    }
 | 
						|
    return dict;
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
parse_item(heim_array_t array, struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    heim_object_t value;
 | 
						|
 | 
						|
    if (white_spaces(ctx))
 | 
						|
	return -1;
 | 
						|
 | 
						|
    if (*ctx->p == ']') {
 | 
						|
	ctx->p++; /* safe because parse_value() calls white_spaces() first */
 | 
						|
	return 0;
 | 
						|
    }
 | 
						|
 | 
						|
    value = parse_value(ctx);
 | 
						|
    if (value == NULL &&
 | 
						|
	(ctx->error || (ctx->flags & HEIM_JSON_F_NO_C_NULL)))
 | 
						|
	return -1;
 | 
						|
 | 
						|
    heim_array_append_value(array, value);
 | 
						|
    heim_release(value);
 | 
						|
 | 
						|
    if (white_spaces(ctx))
 | 
						|
	return -1;
 | 
						|
 | 
						|
    if (*ctx->p == ']') {
 | 
						|
	ctx->p++;
 | 
						|
	return 0;
 | 
						|
    } else if (*ctx->p == ',') {
 | 
						|
	ctx->p++;
 | 
						|
	return 1;
 | 
						|
    }
 | 
						|
    return -1;
 | 
						|
}
 | 
						|
 | 
						|
static heim_array_t
 | 
						|
parse_array(struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    heim_array_t array = heim_array_create();
 | 
						|
    int ret;
 | 
						|
 | 
						|
    heim_assert(*ctx->p == '[', "array doesn't start with [");
 | 
						|
    ctx->p += 1;
 | 
						|
 | 
						|
    while ((ret = parse_item(array, ctx)) > 0)
 | 
						|
	;
 | 
						|
    if (ret < 0) {
 | 
						|
	heim_release(array);
 | 
						|
	return NULL;
 | 
						|
    }
 | 
						|
    return array;
 | 
						|
}
 | 
						|
 | 
						|
static heim_object_t
 | 
						|
parse_value(struct parse_ctx *ctx)
 | 
						|
{
 | 
						|
    size_t len;
 | 
						|
    heim_object_t o;
 | 
						|
 | 
						|
    if (white_spaces(ctx))
 | 
						|
	return NULL;
 | 
						|
 | 
						|
    if (*ctx->p == '"') {
 | 
						|
	return parse_string(ctx);
 | 
						|
    } else if (*ctx->p == '{') {
 | 
						|
	if (ctx->depth-- == 1) {
 | 
						|
	    ctx->error = heim_error_create(EINVAL, "JSON object too deep");
 | 
						|
	    return NULL;
 | 
						|
	}
 | 
						|
	o = parse_dict(ctx);
 | 
						|
	ctx->depth++;
 | 
						|
	return o;
 | 
						|
    } else if (*ctx->p == '[') {
 | 
						|
	if (ctx->depth-- == 1) {
 | 
						|
	    ctx->error = heim_error_create(EINVAL, "JSON object too deep");
 | 
						|
	    return NULL;
 | 
						|
	}
 | 
						|
	o = parse_array(ctx);
 | 
						|
	ctx->depth++;
 | 
						|
	return o;
 | 
						|
    } else if (is_number(*ctx->p) || *ctx->p == '-') {
 | 
						|
	return parse_number(ctx);
 | 
						|
    }
 | 
						|
 | 
						|
    len = ctx->pend - ctx->p;
 | 
						|
 | 
						|
    if ((ctx->flags & HEIM_JSON_F_NO_C_NULL) == 0 &&
 | 
						|
	len >= 6 && memcmp(ctx->p, "<NULL>", 6) == 0) {
 | 
						|
	ctx->p += 6;
 | 
						|
	return heim_null_create();
 | 
						|
    } else if (len >= 4 && memcmp(ctx->p, "null", 4) == 0) {
 | 
						|
	ctx->p += 4;
 | 
						|
	return heim_null_create();
 | 
						|
    } else if (len >= 4 && strncasecmp((char *)ctx->p, "true", 4) == 0) {
 | 
						|
	ctx->p += 4;
 | 
						|
	return heim_bool_create(1);
 | 
						|
    } else if (len >= 5 && strncasecmp((char *)ctx->p, "false", 5) == 0) {
 | 
						|
	ctx->p += 5;
 | 
						|
	return heim_bool_create(0);
 | 
						|
    }
 | 
						|
 | 
						|
    ctx->error = heim_error_create(EINVAL, "unknown char %c at %lu line %lu",
 | 
						|
				   (char)*ctx->p, 
 | 
						|
				   (unsigned long)(ctx->p - ctx->pstart),
 | 
						|
				   ctx->lineno);
 | 
						|
    return NULL;
 | 
						|
}
 | 
						|
 | 
						|
 | 
						|
heim_object_t
 | 
						|
heim_json_create(const char *string, size_t max_depth, heim_json_flags_t flags,
 | 
						|
		 heim_error_t *error)
 | 
						|
{
 | 
						|
    return heim_json_create_with_bytes(string, strlen(string), max_depth, flags,
 | 
						|
				       error);
 | 
						|
}
 | 
						|
 | 
						|
heim_object_t
 | 
						|
heim_json_create_with_bytes(const void *data, size_t length, size_t max_depth,
 | 
						|
			    heim_json_flags_t flags, heim_error_t *error)
 | 
						|
{
 | 
						|
    struct parse_ctx ctx;
 | 
						|
    heim_object_t o;
 | 
						|
 | 
						|
    heim_base_once_f(&heim_json_once, NULL, json_init_once);
 | 
						|
 | 
						|
    ctx.lineno = 1;
 | 
						|
    ctx.p = data;
 | 
						|
    ctx.pstart = data;
 | 
						|
    ctx.pend = ((uint8_t *)data) + length;
 | 
						|
    ctx.error = NULL;
 | 
						|
    ctx.flags = flags;
 | 
						|
    ctx.depth = max_depth;
 | 
						|
 | 
						|
    o = parse_value(&ctx);
 | 
						|
 | 
						|
    if (o == NULL && error) {
 | 
						|
	*error = ctx.error;
 | 
						|
    } else if (ctx.error) {
 | 
						|
	heim_release(ctx.error);
 | 
						|
    }
 | 
						|
 | 
						|
    return o;
 | 
						|
}
 | 
						|
 | 
						|
 | 
						|
static void
 | 
						|
show_printf(void *ctx, const char *str)
 | 
						|
{
 | 
						|
    if (str == NULL)
 | 
						|
	return;
 | 
						|
    fprintf(ctx, "%s", str);
 | 
						|
}
 | 
						|
 | 
						|
/**
 | 
						|
 * Dump a heimbase object to stderr (useful from the debugger!)
 | 
						|
 *
 | 
						|
 * @param obj object to dump using JSON or JSON-like format
 | 
						|
 *
 | 
						|
 * @addtogroup heimbase
 | 
						|
 */
 | 
						|
void
 | 
						|
heim_show(heim_object_t obj)
 | 
						|
{
 | 
						|
    heim_base2json(obj, stderr, HEIM_JSON_F_NO_DATA_DICT, show_printf);
 | 
						|
}
 | 
						|
 | 
						|
static void
 | 
						|
strbuf_add(void *ctx, const char *str)
 | 
						|
{
 | 
						|
    struct heim_strbuf *strbuf = ctx;
 | 
						|
    size_t len;
 | 
						|
 | 
						|
    if (strbuf->enomem)
 | 
						|
	return;
 | 
						|
 | 
						|
    if (str == NULL) {
 | 
						|
	/*
 | 
						|
	 * Eat the last '\n'; this is used when formatting dict pairs
 | 
						|
	 * and array items so that the ',' separating them is never
 | 
						|
	 * preceded by a '\n'.
 | 
						|
	 */
 | 
						|
	if (strbuf->len > 0 && strbuf->str[strbuf->len - 1] == '\n')
 | 
						|
	    strbuf->len--;
 | 
						|
	return;
 | 
						|
    }
 | 
						|
 | 
						|
    len = strlen(str);
 | 
						|
    if ((len + 1) > (strbuf->alloced - strbuf->len)) {
 | 
						|
	size_t new_len = strbuf->alloced + (strbuf->alloced >> 2) + len + 1;
 | 
						|
	char *s;
 | 
						|
 | 
						|
	s = realloc(strbuf->str, new_len);
 | 
						|
	if (s == NULL) {
 | 
						|
	    strbuf->enomem = 1;
 | 
						|
	    return;
 | 
						|
	}
 | 
						|
	strbuf->str = s;
 | 
						|
	strbuf->alloced = new_len;
 | 
						|
    }
 | 
						|
    /* +1 so we copy the NUL */
 | 
						|
    (void) memcpy(strbuf->str + strbuf->len, str, len + 1);
 | 
						|
    strbuf->len += len;
 | 
						|
    if (strbuf->str[strbuf->len - 1] == '\n' && 
 | 
						|
	strbuf->flags & HEIM_JSON_F_ONE_LINE)
 | 
						|
	strbuf->len--;
 | 
						|
}
 | 
						|
 | 
						|
#define STRBUF_INIT_SZ 64
 | 
						|
 | 
						|
heim_string_t
 | 
						|
heim_json_copy_serialize(heim_object_t obj, heim_json_flags_t flags, heim_error_t *error)
 | 
						|
{
 | 
						|
    heim_string_t str;
 | 
						|
    struct heim_strbuf strbuf;
 | 
						|
    int ret;
 | 
						|
 | 
						|
    if (error)
 | 
						|
	*error = NULL;
 | 
						|
 | 
						|
    memset(&strbuf, 0, sizeof (strbuf));
 | 
						|
    strbuf.str = malloc(STRBUF_INIT_SZ);
 | 
						|
    if (strbuf.str == NULL) {
 | 
						|
	if (error)
 | 
						|
	    *error = heim_error_create_enomem();
 | 
						|
	return NULL;
 | 
						|
    }
 | 
						|
    strbuf.len = 0;
 | 
						|
    strbuf.alloced = STRBUF_INIT_SZ;
 | 
						|
    strbuf.str[0] = '\0';
 | 
						|
    strbuf.flags = flags;
 | 
						|
 | 
						|
    ret = heim_base2json(obj, &strbuf, flags, strbuf_add);
 | 
						|
    if (ret || strbuf.enomem) {
 | 
						|
	if (error) {
 | 
						|
	    if (strbuf.enomem || ret == ENOMEM)
 | 
						|
		*error = heim_error_create_enomem();
 | 
						|
	    else
 | 
						|
		*error = heim_error_create(1, "Impossible to JSON-encode "
 | 
						|
					   "object");
 | 
						|
	}
 | 
						|
	free(strbuf.str);
 | 
						|
	return NULL;
 | 
						|
    }
 | 
						|
    if (flags & HEIM_JSON_F_ONE_LINE) {
 | 
						|
	strbuf.flags &= ~HEIM_JSON_F_ONE_LINE;
 | 
						|
	strbuf_add(&strbuf, "\n");
 | 
						|
    }
 | 
						|
    str = heim_string_ref_create(strbuf.str, free);
 | 
						|
    if (str == NULL) {
 | 
						|
	if (error)
 | 
						|
	    *error = heim_error_create_enomem();
 | 
						|
	free(strbuf.str);
 | 
						|
    }
 | 
						|
    return str;
 | 
						|
}
 | 
						|
 | 
						|
struct heim_eq_f_ctx {
 | 
						|
    heim_dict_t other;
 | 
						|
    int ret;
 | 
						|
};
 | 
						|
 | 
						|
static void
 | 
						|
heim_eq_dict_iter_f(heim_object_t key, heim_object_t val, void *d)
 | 
						|
{
 | 
						|
    struct heim_eq_f_ctx *ctx = d;
 | 
						|
    heim_object_t other_val;
 | 
						|
 | 
						|
    if (!ctx->ret)
 | 
						|
        return;
 | 
						|
 | 
						|
    /*
 | 
						|
     * This doesn't work if the key is an array or a dict, which, anyways,
 | 
						|
     * isn't allowed in JSON, though we allow it.
 | 
						|
     */
 | 
						|
    other_val = heim_dict_get_value(ctx->other, key);
 | 
						|
    ctx->ret = heim_json_eq(val, other_val);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
heim_json_eq(heim_object_t a, heim_object_t b)
 | 
						|
{
 | 
						|
    heim_tid_t atid, btid;
 | 
						|
 | 
						|
    if (a == b)
 | 
						|
        return 1;
 | 
						|
    if (a == NULL || b == NULL)
 | 
						|
        return 0;
 | 
						|
    atid = heim_get_tid(a);
 | 
						|
    btid = heim_get_tid(b);
 | 
						|
    if (atid != btid)
 | 
						|
        return 0;
 | 
						|
    switch (atid) {
 | 
						|
    case HEIM_TID_ARRAY: {
 | 
						|
        size_t len = heim_array_get_length(b);
 | 
						|
        size_t i;
 | 
						|
 | 
						|
        if (heim_array_get_length(a) != len)
 | 
						|
            return 0;
 | 
						|
        for (i = 0; i < len; i++) {
 | 
						|
            if (!heim_json_eq(heim_array_get_value(a, i),
 | 
						|
                              heim_array_get_value(b, i)))
 | 
						|
                return 0;
 | 
						|
        }
 | 
						|
        return 1;
 | 
						|
    }
 | 
						|
    case HEIM_TID_DICT: {
 | 
						|
        struct heim_eq_f_ctx ctx;
 | 
						|
 | 
						|
        ctx.other = b;
 | 
						|
        ctx.ret = 1;
 | 
						|
        heim_dict_iterate_f(a, &ctx, heim_eq_dict_iter_f);
 | 
						|
 | 
						|
        if (ctx.ret) {
 | 
						|
            ctx.other = a;
 | 
						|
            heim_dict_iterate_f(b, &ctx, heim_eq_dict_iter_f);
 | 
						|
        }
 | 
						|
        return ctx.ret;
 | 
						|
    }
 | 
						|
    case HEIM_TID_STRING:
 | 
						|
        return strcmp(heim_string_get_utf8(a), heim_string_get_utf8(b)) == 0;
 | 
						|
    case HEIM_TID_DATA: {
 | 
						|
        return heim_data_get_length(a) == heim_data_get_length(b) &&
 | 
						|
               memcmp(heim_data_get_ptr(a), heim_data_get_ptr(b),
 | 
						|
                      heim_data_get_length(a)) == 0;
 | 
						|
    }
 | 
						|
    case HEIM_TID_NUMBER:
 | 
						|
        return heim_number_get_long(a) == heim_number_get_long(b);
 | 
						|
    case HEIM_TID_NULL:
 | 
						|
    case HEIM_TID_BOOL:
 | 
						|
        return heim_bool_val(a) == heim_bool_val(b);
 | 
						|
    default:
 | 
						|
        break;
 | 
						|
    }
 | 
						|
    return 0;
 | 
						|
}
 |