Mercurial > vim

view src/json.c @ 34394:a400c8f9506f v9.1.0123

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
patch 9.1.0123: MS-Windows: system() may deadlock Commit: https://github.com/vim/vim/commit/52ecc76c7fa1865603f27bc838efaeaa03cad77c Author: GuyBrush <miguel.barro@live.com> Date: Wed Feb 21 20:16:38 2024 +0100 patch 9.1.0123: MS-Windows: system() may deadlock Problem: MS-Windows: system() may deadlock when calling binaries that expect stdin Solution: Ignore the SHELL_EXPAND flag (GuyBrush) This happens on binaries that expect stdin. For example: :echo system("xxd") will cause a deadlock. SHELL_EXPAND is a flag devoted to support the linux implementation of the backtick-expansion mechanism. On linux backtic-expansion relies in the function mch_expand_wildchars() (os_unix.c) that delegates on each specific shell (bash, sh, csh, zsh) the expansion. Basically it composes a shell command that does the expansion and redirects the output to a file and call_shell() it. On windows backtick-expansion is performed by Vim itself. On linux SHELL_EXPAND modifies how mch_call_shell_fork() (os_unix.c) works. This function: - relies on posix fork() to spawn a child process to execute a external command. - Child and parent process communicate using pipes (or pseudoterminal if available). User input (type ahead content) is processed in a loop only if !(SHELL_EXPAND || SHELL_COOKED). Though signals are used to detect Ctrl-C in all cases (the input loop is not necessary to interrupt the function). In the backtick-expansion the external command is the shell command that provides the expansion. For the child redirection: - SHELL_EXPAND replaces stdin, stdout & stderr to /dev/null. This is why the shell command composed includes redirection (otherwise output would be lost). - !SHELL_EXPAND replaces stdin, stdout & stderr with the parent created pipes (or pseudoterminal). Note that the use of SIGINT signal prevents mch_call_shell_fork() from hanging vim. On Windows mch_system_piped() (os_win32.c) (which is only used when the GUI is running) mimics mch_call_shell_fork() (os_unix.c). Win32 lacks fork() and relies on CreateProcessW() and only has pipe support (not pseudoterminal) which makes the implementation much different. But, the key idea is that windows lacks signals, the OS provides support for console apps but gvim is not one. The only way of detecting a Ctrl-C is actually processing user input (type ahead content). By ignoring the user input under SHELL_EXPAND the function can hang gvim. Ignoring SHELL_EXPAND flag has no consequence in Windows because as mentioned above it is only meaningful in linux. closes: #13988 Signed-off-by: GuyBrush <miguel.barro@live.com> Signed-off-by: Christian Brabandt <cb@256bit.org>
author Christian Brabandt <cb@256bit.org>
date Wed, 21 Feb 2024 20:30:02 +0100
parents 7d9d2404a3d4
children
line wrap: on
line source

/* vi:set ts=8 sts=4 sw=4 noet:
 *
 * VIM - Vi IMproved	by Bram Moolenaar
 *
 * Do ":help uganda"  in Vim to read copying and usage conditions.
 * Do ":help credits" in Vim to see a list of people who contributed.
 * See README.txt for an overview of the Vim source code.
 */

/*
 * json.c: Encoding and decoding JSON.
 *
 * Follows this standard: https://tools.ietf.org/html/rfc7159.html
 */
#define USING_FLOAT_STUFF

#include "vim.h"

#if defined(FEAT_EVAL) || defined(PROTO)

static int json_encode_item(garray_T *gap, typval_T *val, int copyID, int options);

/*
 * Encode "val" into a JSON format string.
 * The result is added to "gap"
 * Returns FAIL on failure and makes gap->ga_data empty.
 */
    static int
json_encode_gap(garray_T *gap, typval_T *val, int options)
{
    if (json_encode_item(gap, val, get_copyID(), options) == FAIL)
    {
	ga_clear(gap);
	gap->ga_data = vim_strsave((char_u *)"");
	return FAIL;
    }
    return OK;
}

/*
 * Encode "val" into a JSON format string.
 * The result is in allocated memory.
 * The result is empty when encoding fails.
 * "options" can contain JSON_JS, JSON_NO_NONE and JSON_NL.
 */
    char_u *
json_encode(typval_T *val, int options)
{
    garray_T ga;

    // Store bytes in the growarray.
    ga_init2(&ga, 1, 4000);
    json_encode_gap(&ga, val, options);
    ga_append(&ga, NUL);
    return ga.ga_data;
}

#if defined(FEAT_JOB_CHANNEL) || defined(PROTO)
/*
 * Encode ["nr", "val"] into a JSON format string in allocated memory.
 * "options" can contain JSON_JS, JSON_NO_NONE and JSON_NL.
 * Returns NULL when out of memory.
 */
    char_u *
json_encode_nr_expr(int nr, typval_T *val, int options)
{
    typval_T	listtv;
    typval_T	nrtv;
    garray_T	ga;

    nrtv.v_type = VAR_NUMBER;
    nrtv.vval.v_number = nr;
    if (rettv_list_alloc(&listtv) == FAIL)
	return NULL;
    if (list_append_tv(listtv.vval.v_list, &nrtv) == FAIL
	    || list_append_tv(listtv.vval.v_list, val) == FAIL)
    {
	list_unref(listtv.vval.v_list);
	return NULL;
    }

    ga_init2(&ga, 1, 4000);
    if (json_encode_gap(&ga, &listtv, options) == OK && (options & JSON_NL))
	ga_append(&ga, '\n');
    list_unref(listtv.vval.v_list);
    ga_append(&ga, NUL);
    return ga.ga_data;
}

/*
 * Encode "val" into a JSON format string prefixed by the LSP HTTP header.
 * Returns NULL when out of memory.
 */
    char_u *
json_encode_lsp_msg(typval_T *val)
{
    garray_T	ga;
    garray_T	lspga;

    ga_init2(&ga, 1, 4000);
    if (json_encode_gap(&ga, val, 0) == FAIL)
	return NULL;
    ga_append(&ga, NUL);

    ga_init2(&lspga, 1, 4000);
    // Header according to LSP specification.
    vim_snprintf((char *)IObuff, IOSIZE,
	    "Content-Length: %u\r\n\r\n",
	    ga.ga_len - 1);
    ga_concat(&lspga, IObuff);
    ga_concat_len(&lspga, ga.ga_data, ga.ga_len);
    ga_clear(&ga);
    return lspga.ga_data;
}
#endif

/*
 * Lookup table to quickly know if the given ASCII character must be escaped.
 */
static const char ascii_needs_escape[128] = {
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 0x0.
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 0x1.
    0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x2.
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x3.
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x4.
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, // 0x5.
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x6.
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x7.
};

/*
 * Encode the utf-8 encoded string "str" into "gap".
 */
    static void
write_string(garray_T *gap, char_u *str)
{
    char_u	*res = str;
    char_u	numbuf[NUMBUFLEN];
    char_u	*from;
#if defined(USE_ICONV)
    vimconv_T   conv;
    char_u	*converted = NULL;
#endif
    int		c;

    if (res == NULL)
    {
	ga_concat(gap, (char_u *)"\"\"");
	return;
    }

#if defined(USE_ICONV)
    if (!enc_utf8)
    {
	// Convert the text from 'encoding' to utf-8, because a JSON string is
	// always utf-8.
	conv.vc_type = CONV_NONE;
	convert_setup(&conv, p_enc, (char_u*)"utf-8");
	if (conv.vc_type != CONV_NONE)
	    converted = res = string_convert(&conv, res, NULL);
	convert_setup(&conv, NULL, NULL);
    }
#endif
    ga_append(gap, '"');
    // `from` is the beginning of a sequence of bytes we can directly copy from
    // the input string, avoiding the overhead associated to decoding/encoding
    // them.
    from = res;
    while ((c = *res) != NUL)
    {
	// always use utf-8 encoding, ignore 'encoding'
	if (c < 0x80)
	{
	    if (!ascii_needs_escape[c])
	    {
		res += 1;
		continue;
	    }

	    if (res != from)
		ga_concat_len(gap, from, res - from);
	    from = res + 1;

	    switch (c)
	    {
		case 0x08:
		    ga_append(gap, '\\'); ga_append(gap, 'b'); break;
		case 0x09:
		    ga_append(gap, '\\'); ga_append(gap, 't'); break;
		case 0x0a:
		    ga_append(gap, '\\'); ga_append(gap, 'n'); break;
		case 0x0c:
		    ga_append(gap, '\\'); ga_append(gap, 'f'); break;
		case 0x0d:
		    ga_append(gap, '\\'); ga_append(gap, 'r'); break;
		case 0x22: // "
		case 0x5c: // backslash
		    ga_append(gap, '\\');
		    ga_append(gap, c);
		    break;
		default:
		    vim_snprintf((char *)numbuf, NUMBUFLEN, "\\u%04lx",
								      (long)c);
		    ga_concat(gap, numbuf);
	    }

	    res += 1;
	}
	else
	{
	    int l = utf_ptr2len(res);

	    if (l > 1)
	    {
		res += l;
		continue;
	    }

	    // Invalid utf-8 sequence, replace it with the Unicode replacement
	    // character U+FFFD.
	    if (res != from)
		ga_concat_len(gap, from, res - from);
	    from = res + 1;

	    numbuf[utf_char2bytes(0xFFFD, numbuf)] = NUL;
	    ga_concat(gap, numbuf);

	    res += l;
	}
    }

    if (res != from)
	ga_concat_len(gap, from, res - from);

    ga_append(gap, '"');
#if defined(USE_ICONV)
    vim_free(converted);
#endif
}

/*
 * Return TRUE if "key" can be used without quotes.
 * That is when it starts with a letter and only contains letters, digits and
 * underscore.
 */
    static int
is_simple_key(char_u *key)
{
    char_u *p;

    if (!ASCII_ISALPHA(*key))
	return FALSE;
    for (p = key + 1; *p != NUL; ++p)
	if (!ASCII_ISALPHA(*p) && *p != '_' && !vim_isdigit(*p))
	    return FALSE;
    return TRUE;
}

/*
 * Encode "val" into "gap".
 * Return FAIL or OK.
 */
    static int
json_encode_item(garray_T *gap, typval_T *val, int copyID, int options)
{
    char_u	numbuf[NUMBUFLEN];
    char_u	*res;
    blob_T	*b;
    list_T	*l;
    dict_T	*d;
    int		i;

    switch (val->v_type)
    {
	case VAR_BOOL:
	    switch ((long)val->vval.v_number)
	    {
		case VVAL_FALSE: ga_concat(gap, (char_u *)"false"); break;
		case VVAL_TRUE: ga_concat(gap, (char_u *)"true"); break;
	    }
	    break;

	case VAR_SPECIAL:
	    switch ((long)val->vval.v_number)
	    {
		case VVAL_NONE: if ((options & JSON_JS) != 0
					     && (options & JSON_NO_NONE) == 0)
				    // empty item
				    break;
				// FALLTHROUGH
		case VVAL_NULL: ga_concat(gap, (char_u *)"null"); break;
	    }
	    break;

	case VAR_NUMBER:
	    vim_snprintf((char *)numbuf, NUMBUFLEN, "%lld",
					      (varnumber_T)val->vval.v_number);
	    ga_concat(gap, numbuf);
	    break;

	case VAR_STRING:
	    res = val->vval.v_string;
	    write_string(gap, res);
	    break;

	case VAR_FUNC:
	case VAR_PARTIAL:
	case VAR_JOB:
	case VAR_CHANNEL:
	case VAR_INSTR:
	case VAR_CLASS:
	case VAR_OBJECT:
	case VAR_TYPEALIAS:
	    semsg(_(e_cannot_json_encode_str), vartype_name(val->v_type));
	    return FAIL;

	case VAR_BLOB:
	    b = val->vval.v_blob;
	    if (b == NULL || b->bv_ga.ga_len == 0)
		ga_concat(gap, (char_u *)"[]");
	    else
	    {
		ga_append(gap, '[');
		for (i = 0; i < b->bv_ga.ga_len; i++)
		{
		    if (i > 0)
			ga_concat(gap, (char_u *)",");
		    vim_snprintf((char *)numbuf, NUMBUFLEN, "%d",
			    blob_get(b, i));
		    ga_concat(gap, numbuf);
		}
		ga_append(gap, ']');
	    }
	    break;

	case VAR_LIST:
	    l = val->vval.v_list;
	    if (l == NULL)
		ga_concat(gap, (char_u *)"[]");
	    else
	    {
		if (l->lv_copyID == copyID)
		    ga_concat(gap, (char_u *)"[]");
		else
		{
		    listitem_T	*li;

		    l->lv_copyID = copyID;
		    ga_append(gap, '[');
		    CHECK_LIST_MATERIALIZE(l);
		    for (li = l->lv_first; li != NULL && !got_int; )
		    {
			if (json_encode_item(gap, &li->li_tv, copyID,
						   options & JSON_JS) == FAIL)
			    return FAIL;
			if ((options & JSON_JS)
				&& li->li_next == NULL
				&& li->li_tv.v_type == VAR_SPECIAL
				&& li->li_tv.vval.v_number == VVAL_NONE)
			    // add an extra comma if the last item is v:none
			    ga_append(gap, ',');
			li = li->li_next;
			if (li != NULL)
			    ga_append(gap, ',');
		    }
		    ga_append(gap, ']');
		    l->lv_copyID = 0;
		}
	    }
	    break;

	case VAR_DICT:
	    d = val->vval.v_dict;
	    if (d == NULL)
		ga_concat(gap, (char_u *)"{}");
	    else
	    {
		if (d->dv_copyID == copyID)
		    ga_concat(gap, (char_u *)"{}");
		else
		{
		    int		first = TRUE;
		    int		todo = (int)d->dv_hashtab.ht_used;
		    hashitem_T	*hi;

		    d->dv_copyID = copyID;
		    ga_append(gap, '{');

		    for (hi = d->dv_hashtab.ht_array; todo > 0 && !got_int;
									 ++hi)
			if (!HASHITEM_EMPTY(hi))
			{
			    --todo;
			    if (first)
				first = FALSE;
			    else
				ga_append(gap, ',');
			    if ((options & JSON_JS)
						 && is_simple_key(hi->hi_key))
				ga_concat(gap, hi->hi_key);
			    else
				write_string(gap, hi->hi_key);
			    ga_append(gap, ':');
			    if (json_encode_item(gap, &dict_lookup(hi)->di_tv,
				      copyID, options | JSON_NO_NONE) == FAIL)
				return FAIL;
			}
		    ga_append(gap, '}');
		    d->dv_copyID = 0;
		}
	    }
	    break;

	case VAR_FLOAT:
#if defined(HAVE_MATH_H)
	    if (isnan(val->vval.v_float))
		ga_concat(gap, (char_u *)"NaN");
	    else if (isinf(val->vval.v_float))
	    {
		if (val->vval.v_float < 0.0)
		    ga_concat(gap, (char_u *)"-Infinity");
		else
		    ga_concat(gap, (char_u *)"Infinity");
	    }
	    else
#endif
	    {
		vim_snprintf((char *)numbuf, NUMBUFLEN, "%g",
							   val->vval.v_float);
		ga_concat(gap, numbuf);
	    }
	    break;
	case VAR_UNKNOWN:
	case VAR_ANY:
	case VAR_VOID:
	    internal_error_no_abort("json_encode_item()");
	    return FAIL;
    }
    return OK;
}

/*
 * When "reader" has less than NUMBUFLEN bytes available, call the fill
 * callback to get more.
 */
    static void
fill_numbuflen(js_read_T *reader)
{
    if (reader->js_fill != NULL && (int)(reader->js_end - reader->js_buf)
						- reader->js_used < NUMBUFLEN)
    {
	if (reader->js_fill(reader))
	    reader->js_end = reader->js_buf + STRLEN(reader->js_buf);
    }
}

/*
 * Skip white space in "reader".  All characters <= space are considered white
 * space.
 * Also tops up readahead when needed.
 */
    static void
json_skip_white(js_read_T *reader)
{
    int c;

    for (;;)
    {
	c = reader->js_buf[reader->js_used];
	if (reader->js_fill != NULL && c == NUL)
	{
	    if (reader->js_fill(reader))
	    {
		reader->js_end = reader->js_buf + STRLEN(reader->js_buf);
		continue;
	    }
	}
	if (c == NUL || c > ' ')
	    break;
	++reader->js_used;
    }
    fill_numbuflen(reader);
}

    static int
json_decode_string(js_read_T *reader, typval_T *res, int quote)
{
    garray_T    ga;
    int		len;
    char_u	*p;
    int		c;
    varnumber_T	nr;

    if (res != NULL)
	ga_init2(&ga, 1, 200);

    p = reader->js_buf + reader->js_used + 1; // skip over " or '
    while (*p != quote)
    {
	// The JSON is always expected to be utf-8, thus use utf functions
	// here. The string is converted below if needed.
	if (*p == NUL || p[1] == NUL || utf_ptr2len(p) < utf_byte2len(*p))
	{
	    // Not enough bytes to make a character or end of the string. Get
	    // more if possible.
	    if (reader->js_fill == NULL)
		break;
	    len = (int)(reader->js_end - p);
	    reader->js_used = (int)(p - reader->js_buf);
	    if (!reader->js_fill(reader))
		break; // didn't get more
	    p = reader->js_buf + reader->js_used;
	    reader->js_end = reader->js_buf + STRLEN(reader->js_buf);
	    continue;
	}

	if (*p == '\\')
	{
	    c = -1;
	    switch (p[1])
	    {
		case '\\': c = '\\'; break;
		case '"': c = '"'; break;
		case 'b': c = BS; break;
		case 't': c = TAB; break;
		case 'n': c = NL; break;
		case 'f': c = FF; break;
		case 'r': c = CAR; break;
		case 'u':
		    if (reader->js_fill != NULL
				     && (int)(reader->js_end - p) < NUMBUFLEN)
		    {
			reader->js_used = (int)(p - reader->js_buf);
			if (reader->js_fill(reader))
			{
			    p = reader->js_buf + reader->js_used;
			    reader->js_end = reader->js_buf
						     + STRLEN(reader->js_buf);
			}
		    }
		    nr = 0;
		    len = 0;
		    vim_str2nr(p + 2, NULL, &len,
			     STR2NR_HEX + STR2NR_FORCE, &nr, NULL, 4, TRUE, NULL);
		    if (len == 0)
		    {
			if (res != NULL)
			    ga_clear(&ga);
			return FAIL;
		    }
		    p += len + 2;
		    if (0xd800 <= nr && nr <= 0xdfff
			    && (int)(reader->js_end - p) >= 6
			    && *p == '\\' && *(p+1) == 'u')
		    {
			varnumber_T	nr2 = 0;

			// decode surrogate pair: \ud812\u3456
			len = 0;
			vim_str2nr(p + 2, NULL, &len, STR2NR_HEX + STR2NR_FORCE,
						    &nr2, NULL, 4, TRUE, NULL);
			if (len == 0)
			{
			    if (res != NULL)
				ga_clear(&ga);
			    return FAIL;
			}
			if (0xdc00 <= nr2 && nr2 <= 0xdfff)
			{
			    p += len + 2;
			    nr = (((nr - 0xd800) << 10) |
				((nr2 - 0xdc00) & 0x3ff)) + 0x10000;
			}
		    }
		    if (res != NULL)
		    {
			char_u	buf[NUMBUFLEN];

			buf[utf_char2bytes((int)nr, buf)] = NUL;
			ga_concat(&ga, buf);
		    }
		    break;
		default:
		    // not a special char, skip over backslash
		    ++p;
		    continue;
	    }
	    if (c > 0)
	    {
		p += 2;
		if (res != NULL)
		    ga_append(&ga, c);
	    }
	}
	else
	{
	    len = utf_ptr2len(p);
	    if (res != NULL)
	    {
		if (ga_grow(&ga, len) == FAIL)
		{
		    ga_clear(&ga);
		    return FAIL;
		}
		mch_memmove((char *)ga.ga_data + ga.ga_len, p, (size_t)len);
		ga.ga_len += len;
	    }
	    p += len;
	}
    }

    reader->js_used = (int)(p - reader->js_buf);
    if (*p == quote)
    {
	++reader->js_used;
	if (res != NULL)
	{
	    ga_append(&ga, NUL);
	    res->v_type = VAR_STRING;
#if defined(USE_ICONV)
	    if (!enc_utf8)
	    {
		vimconv_T   conv;

		// Convert the utf-8 string to 'encoding'.
		conv.vc_type = CONV_NONE;
		convert_setup(&conv, (char_u*)"utf-8", p_enc);
		if (conv.vc_type != CONV_NONE)
		{
		    res->vval.v_string =
				      string_convert(&conv, ga.ga_data, NULL);
		    vim_free(ga.ga_data);
		}
		convert_setup(&conv, NULL, NULL);
	    }
	    else
#endif
		res->vval.v_string = ga.ga_data;
	}
	return OK;
    }
    if (res != NULL)
    {
	res->v_type = VAR_SPECIAL;
	res->vval.v_number = VVAL_NONE;
	ga_clear(&ga);
    }
    return MAYBE;
}

typedef enum {
    JSON_ARRAY,		// parsing items in an array
    JSON_OBJECT_KEY,	// parsing key of an object
    JSON_OBJECT		// parsing item in an object, after the key
} json_decode_T;

typedef struct {
    json_decode_T jd_type;
    typval_T	  jd_tv;	// the list or dict
    typval_T	  jd_key_tv;
    char_u	  *jd_key;
} json_dec_item_T;

/*
 * Decode one item and put it in "res".  If "res" is NULL only advance.
 * Must already have skipped white space.
 *
 * Return FAIL for a decoding error (and give an error).
 * Return MAYBE for an incomplete message.
 */
    static int
json_decode_item(js_read_T *reader, typval_T *res, int options)
{
    char_u	*p;
    int		i;
    int		len;
    int		retval;
    garray_T	stack;
    typval_T	item;
    typval_T	*cur_item;
    json_dec_item_T *top_item;
    char_u	key_buf[NUMBUFLEN];

    ga_init2(&stack, sizeof(json_dec_item_T), 100);
    cur_item = res;
    init_tv(&item);
    if (res != NULL)
	init_tv(res);

    fill_numbuflen(reader);
    p = reader->js_buf + reader->js_used;
    for (;;)
    {
	top_item = NULL;
	if (stack.ga_len > 0)
	{
	    top_item = ((json_dec_item_T *)stack.ga_data) + stack.ga_len - 1;
	    json_skip_white(reader);
	    p = reader->js_buf + reader->js_used;
	    if (*p == NUL)
	    {
		retval = MAYBE;
		goto theend;
	    }
	    if (top_item->jd_type == JSON_OBJECT_KEY
					    || top_item->jd_type == JSON_ARRAY)
	    {
		// Check for end of object or array.
		if (*p == (top_item->jd_type == JSON_ARRAY ? ']' : '}'))
		{
		    ++reader->js_used; // consume the ']' or '}'
		    --stack.ga_len;
		    if (stack.ga_len == 0)
		    {
			retval = OK;
			goto theend;
		    }
		    if (cur_item != NULL)
			cur_item = &top_item->jd_tv;
		    goto item_end;
		}
	    }
	}

	if (top_item != NULL && top_item->jd_type == JSON_OBJECT_KEY
		&& (options & JSON_JS)
		&& reader->js_buf[reader->js_used] != '"'
		&& reader->js_buf[reader->js_used] != '\''
		&& reader->js_buf[reader->js_used] != '['
		&& reader->js_buf[reader->js_used] != '{')
	{
	    char_u *key;

	    // accept an object key that is not in quotes
	    key = p = reader->js_buf + reader->js_used;
	    while (*p != NUL && *p != ':' && *p > ' ')
		++p;
	    if (cur_item != NULL)
	    {
		cur_item->v_type = VAR_STRING;
		cur_item->vval.v_string = vim_strnsave(key, p - key);
		top_item->jd_key = cur_item->vval.v_string;
	    }
	    reader->js_used += (int)(p - key);
	}
	else
	{
	    switch (*p)
	    {
		case '[': // start of array
		    if (top_item && top_item->jd_type == JSON_OBJECT_KEY)
		    {
			retval = FAIL;
			break;
		    }
		    if (ga_grow(&stack, 1) == FAIL)
		    {
			retval = FAIL;
			break;
		    }
		    if (cur_item != NULL && rettv_list_alloc(cur_item) == FAIL)
		    {
			cur_item->v_type = VAR_SPECIAL;
			cur_item->vval.v_number = VVAL_NONE;
			retval = FAIL;
			break;
		    }

		    ++reader->js_used; // consume the '['
		    top_item = ((json_dec_item_T *)stack.ga_data)
								+ stack.ga_len;
		    top_item->jd_type = JSON_ARRAY;
		    ++stack.ga_len;
		    if (cur_item != NULL)
		    {
			top_item->jd_tv = *cur_item;
			cur_item = &item;
		    }
		    continue;

		case '{': // start of object
		    if (top_item && top_item->jd_type == JSON_OBJECT_KEY)
		    {
			retval = FAIL;
			break;
		    }
		    if (ga_grow(&stack, 1) == FAIL)
		    {
			retval = FAIL;
			break;
		    }
		    if (cur_item != NULL && rettv_dict_alloc(cur_item) == FAIL)
		    {
			cur_item->v_type = VAR_SPECIAL;
			cur_item->vval.v_number = VVAL_NONE;
			retval = FAIL;
			break;
		    }

		    ++reader->js_used; // consume the '{'
		    top_item = ((json_dec_item_T *)stack.ga_data)
								+ stack.ga_len;
		    top_item->jd_type = JSON_OBJECT_KEY;
		    ++stack.ga_len;
		    if (cur_item != NULL)
		    {
			top_item->jd_tv = *cur_item;
			cur_item = &top_item->jd_key_tv;
		    }
		    continue;

		case '"': // string
		    retval = json_decode_string(reader, cur_item, *p);
		    break;

		case '\'':
		    if (options & JSON_JS)
			retval = json_decode_string(reader, cur_item, *p);
		    else
		    {
			semsg(_(e_json_decode_error_at_str), p);
			retval = FAIL;
		    }
		    break;

		case ',': // comma: empty item
		    if ((options & JSON_JS) == 0)
		    {
			semsg(_(e_json_decode_error_at_str), p);
			retval = FAIL;
			break;
		    }
		    // FALLTHROUGH
		case NUL: // empty
		    if (cur_item != NULL)
		    {
			cur_item->v_type = VAR_SPECIAL;
			cur_item->vval.v_number = VVAL_NONE;
		    }
		    retval = OK;
		    break;

		default:
		    if (VIM_ISDIGIT(*p) || (*p == '-'
					&& (VIM_ISDIGIT(p[1]) || p[1] == NUL)))
		    {
			char_u  *sp = p;

			if (*sp == '-')
			{
			    ++sp;
			    if (*sp == NUL)
			    {
				retval = MAYBE;
				break;
			    }
			    if (!VIM_ISDIGIT(*sp))
			    {
				semsg(_(e_json_decode_error_at_str), p);
				retval = FAIL;
				break;
			    }
			}
			sp = skipdigits(sp);
			if (*sp == '.' || *sp == 'e' || *sp == 'E')
			{
			    if (cur_item == NULL)
			    {
				float_T f;

				len = string2float(p, &f, FALSE);
			    }
			    else
			    {
				cur_item->v_type = VAR_FLOAT;
				len = string2float(p, &cur_item->vval.v_float,
									FALSE);
			    }
			}
			else
			{
			    varnumber_T nr;

			    vim_str2nr(reader->js_buf + reader->js_used,
				    NULL, &len, 0, // what
				    &nr, NULL, 0, TRUE, NULL);
			    if (len == 0)
			    {
				semsg(_(e_json_decode_error_at_str), p);
				retval = FAIL;
				goto theend;
			    }
			    if (cur_item != NULL)
			    {
				cur_item->v_type = VAR_NUMBER;
				cur_item->vval.v_number = nr;
			    }
			}
			reader->js_used += len;
			retval = OK;
			break;
		    }
		    if (STRNICMP((char *)p, "false", 5) == 0)
		    {
			reader->js_used += 5;
			if (cur_item != NULL)
			{
			    cur_item->v_type = VAR_BOOL;
			    cur_item->vval.v_number = VVAL_FALSE;
			}
			retval = OK;
			break;
		    }
		    if (STRNICMP((char *)p, "true", 4) == 0)
		    {
			reader->js_used += 4;
			if (cur_item != NULL)
			{
			    cur_item->v_type = VAR_BOOL;
			    cur_item->vval.v_number = VVAL_TRUE;
			}
			retval = OK;
			break;
		    }
		    if (STRNICMP((char *)p, "null", 4) == 0)
		    {
			reader->js_used += 4;
			if (cur_item != NULL)
			{
			    cur_item->v_type = VAR_SPECIAL;
			    cur_item->vval.v_number = VVAL_NULL;
			}
			retval = OK;
			break;
		    }
		    if (STRNICMP((char *)p, "NaN", 3) == 0)
		    {
			reader->js_used += 3;
			if (cur_item != NULL)
			{
			    cur_item->v_type = VAR_FLOAT;
			    cur_item->vval.v_float = NAN;
			}
			retval = OK;
			break;
		    }
		    if (STRNICMP((char *)p, "-Infinity", 9) == 0)
		    {
			reader->js_used += 9;
			if (cur_item != NULL)
			{
			    cur_item->v_type = VAR_FLOAT;
			    cur_item->vval.v_float = -INFINITY;
			}
			retval = OK;
			break;
		    }
		    if (STRNICMP((char *)p, "Infinity", 8) == 0)
		    {
			reader->js_used += 8;
			if (cur_item != NULL)
			{
			    cur_item->v_type = VAR_FLOAT;
			    cur_item->vval.v_float = INFINITY;
			}
			retval = OK;
			break;
		    }
		    // check for truncated name
		    len = (int)(reader->js_end
					 - (reader->js_buf + reader->js_used));
		    if (
			    (len < 5 && STRNICMP((char *)p, "false", len) == 0)
			    || (len < 9
				 && STRNICMP((char *)p, "-Infinity", len) == 0)
			    || (len < 8
				  && STRNICMP((char *)p, "Infinity", len) == 0)
			    || (len < 3 && STRNICMP((char *)p, "NaN", len) == 0)
			    || (len < 4
				  && (STRNICMP((char *)p, "true", len) == 0
				    || STRNICMP((char *)p, "null", len) == 0)))

			retval = MAYBE;
		    else
			retval = FAIL;
		    break;
	    }

	    // We are finished when retval is FAIL or MAYBE and when at the
	    // toplevel.
	    if (retval == FAIL)
		break;
	    if (retval == MAYBE || stack.ga_len == 0)
		goto theend;

	    if (top_item != NULL && top_item->jd_type == JSON_OBJECT_KEY
		    && cur_item != NULL)
	    {
		if (cur_item->v_type == VAR_FLOAT)
		{
		    // cannot use a float as a key
		    emsg(_(e_using_float_as_string));
		    retval = FAIL;
		    goto theend;
		}
		top_item->jd_key = tv_get_string_buf_chk(cur_item, key_buf);
		if (top_item->jd_key == NULL)
		{
		    emsg(_(e_invalid_argument));
		    retval = FAIL;
		    goto theend;
		}
	    }
	}

item_end:
	top_item = ((json_dec_item_T *)stack.ga_data) + stack.ga_len - 1;
	switch (top_item->jd_type)
	{
	    case JSON_ARRAY:
		if (res != NULL)
		{
		    listitem_T	*li = listitem_alloc();

		    if (li == NULL)
		    {
			clear_tv(cur_item);
			retval = FAIL;
			goto theend;
		    }
		    li->li_tv = *cur_item;
		    list_append(top_item->jd_tv.vval.v_list, li);
		}
		if (cur_item != NULL)
		    cur_item = &item;

		json_skip_white(reader);
		p = reader->js_buf + reader->js_used;
		if (*p == ',')
		    ++reader->js_used;
		else if (*p != ']')
		{
		    if (*p == NUL)
			retval = MAYBE;
		    else
		    {
			semsg(_(e_json_decode_error_at_str), p);
			retval = FAIL;
		    }
		    goto theend;
		}
		break;

	    case JSON_OBJECT_KEY:
		json_skip_white(reader);
		p = reader->js_buf + reader->js_used;
		if (*p != ':')
		{
		    if (cur_item != NULL)
			clear_tv(cur_item);
		    if (*p == NUL)
			retval = MAYBE;
		    else
		    {
			semsg(_(e_json_decode_error_at_str), p);
			retval = FAIL;
		    }
		    goto theend;
		}
		++reader->js_used;
		json_skip_white(reader);
		top_item->jd_type = JSON_OBJECT;
		if (cur_item != NULL)
		    cur_item = &item;
		break;

	    case JSON_OBJECT:
		if (cur_item != NULL
			&& dict_has_key(top_item->jd_tv.vval.v_dict,
						(char *)top_item->jd_key))
		{
		    semsg(_(e_duplicate_key_in_json_str), top_item->jd_key);
		    clear_tv(cur_item);
		    retval = FAIL;
		    goto theend;
		}

		if (cur_item != NULL)
		{
		    dictitem_T *di = dictitem_alloc(top_item->jd_key);

		    clear_tv(&top_item->jd_key_tv);
		    if (di == NULL)
		    {
			clear_tv(cur_item);
			retval = FAIL;
			goto theend;
		    }
		    di->di_tv = *cur_item;
		    di->di_tv.v_lock = 0;
		    if (dict_add(top_item->jd_tv.vval.v_dict, di) == FAIL)
		    {
			dictitem_free(di);
			retval = FAIL;
			goto theend;
		    }
		}

		json_skip_white(reader);
		p = reader->js_buf + reader->js_used;
		if (*p == ',')
		    ++reader->js_used;
		else if (*p != '}')
		{
		    if (*p == NUL)
			retval = MAYBE;
		    else
		    {
			semsg(_(e_json_decode_error_at_str), p);
			retval = FAIL;
		    }
		    goto theend;
		}
		top_item->jd_type = JSON_OBJECT_KEY;
		if (cur_item != NULL)
		    cur_item = &top_item->jd_key_tv;
		break;
	}
    }

    // Get here when parsing failed.
    if (res != NULL)
    {
	clear_tv(res);
	res->v_type = VAR_SPECIAL;
	res->vval.v_number = VVAL_NONE;
    }
    semsg(_(e_json_decode_error_at_str), p);

theend:
    for (i = 0; i < stack.ga_len; i++)
	clear_tv(&(((json_dec_item_T *)stack.ga_data) + i)->jd_key_tv);
    ga_clear(&stack);

    return retval;
}

/*
 * Decode the JSON from "reader" and store the result in "res".
 * "options" can be JSON_JS or zero;
 * Return FAIL if not the whole message was consumed.
 */
    static int
json_decode_all(js_read_T *reader, typval_T *res, int options)
{
    int ret;

    // We find the end once, to avoid calling strlen() many times.
    reader->js_end = reader->js_buf + STRLEN(reader->js_buf);
    json_skip_white(reader);
    ret = json_decode_item(reader, res, options);
    if (ret != OK)
    {
	if (ret == MAYBE)
	    semsg(_(e_json_decode_error_at_str), reader->js_buf);
	return FAIL;
    }
    json_skip_white(reader);
    if (reader->js_buf[reader->js_used] != NUL)
    {
	semsg(_(e_trailing_characters_str), reader->js_buf + reader->js_used);
	return FAIL;
    }
    return OK;
}

#if defined(FEAT_JOB_CHANNEL) || defined(PROTO)
/*
 * Decode the JSON from "reader" and store the result in "res".
 * "options" can be JSON_JS or zero;
 * Return FAIL for a decoding error.
 * Return MAYBE for an incomplete message.
 * Consumes the message anyway.
 */
    int
json_decode(js_read_T *reader, typval_T *res, int options)
{
    int ret;

    // We find the end once, to avoid calling strlen() many times.
    reader->js_end = reader->js_buf + STRLEN(reader->js_buf);
    json_skip_white(reader);
    ret = json_decode_item(reader, res, options);
    json_skip_white(reader);

    return ret;
}
#endif

/*
 * Decode the JSON from "reader" to find the end of the message.
 * "options" can be JSON_JS or zero.
 * This is only used for testing.
 * Return FAIL if the message has a decoding error.
 * Return MAYBE if the message is truncated, need to read more.
 * This only works reliable if the message contains an object, array or
 * string.  A number might be truncated without knowing.
 * Does not advance the reader.
 */
    int
json_find_end(js_read_T *reader, int options)
{
    int used_save = reader->js_used;
    int ret;

    // We find the end once, to avoid calling strlen() many times.
    reader->js_end = reader->js_buf + STRLEN(reader->js_buf);
    json_skip_white(reader);
    ret = json_decode_item(reader, NULL, options);
    reader->js_used = used_save;
    return ret;
}

/*
 * "js_decode()" function
 */
    void
f_js_decode(typval_T *argvars, typval_T *rettv)
{
    js_read_T	reader;

    if (in_vim9script() && check_for_string_arg(argvars, 0) == FAIL)
	return;

    reader.js_buf = tv_get_string(&argvars[0]);
    reader.js_fill = NULL;
    reader.js_used = 0;
    if (json_decode_all(&reader, rettv, JSON_JS) != OK)
	emsg(_(e_invalid_argument));
}

/*
 * "js_encode()" function
 */
    void
f_js_encode(typval_T *argvars, typval_T *rettv)
{
    rettv->v_type = VAR_STRING;
    rettv->vval.v_string = json_encode(&argvars[0], JSON_JS);
}

/*
 * "json_decode()" function
 */
    void
f_json_decode(typval_T *argvars, typval_T *rettv)
{
    js_read_T	reader;

    if (in_vim9script() && check_for_string_arg(argvars, 0) == FAIL)
	return;

    reader.js_buf = tv_get_string(&argvars[0]);
    reader.js_fill = NULL;
    reader.js_used = 0;
    json_decode_all(&reader, rettv, 0);
}

/*
 * "json_encode()" function
 */
    void
f_json_encode(typval_T *argvars, typval_T *rettv)
{
    rettv->v_type = VAR_STRING;
    rettv->vval.v_string = json_encode(&argvars[0], 0);
}
#endif