mirror of
git://git.psyced.org/git/psyced
synced 2024-08-15 03:25:10 +00:00
629 lines
12 KiB
Text
629 lines
12 KiB
Text
// vim:syntax=lpc
|
|
// $Id: jsonparser.pike,v 1.5 2008/08/05 12:21:34 lynx Exp $
|
|
//
|
|
// I really hate those comments.
|
|
//
|
|
// Copyright History:
|
|
// 1. Public Domain 2002 JSON.org
|
|
// 2. Ported to C# by Are Bjolseth, teleplan.no
|
|
// 3. Ported to pike by Bill Welliver
|
|
// "Last night, I downloaded the JSON-C# code, and converted it to pike
|
|
// (with relatively little effort, it was mostly a tedious reformatting
|
|
// job)."
|
|
// 4. Adopted by Tobias 'tobij' Josefowitz, now uses Pike datastructures,
|
|
// and can only be run in LDMud.
|
|
//
|
|
// As far as I am concerned, this still is Public Domain.
|
|
// I will probably once find out whether Are and Bill think the same wwy
|
|
// about it.
|
|
|
|
#ifdef __PIKE__
|
|
/// <summary>
|
|
/// <para>
|
|
/// A JSONTokener takes a source string and extracts characters and tokens from
|
|
/// it. It is used by the JSONObject and JSONArray constructors to parse
|
|
/// JSON source strings.
|
|
/// </para>
|
|
/// <para>
|
|
/// Public Domain 2002 JSON.org
|
|
/// @author JSON.org
|
|
/// @version 0.1
|
|
/// </para>
|
|
/// <para>Ported to C# by Are Bjolseth, teleplan.no</para>
|
|
/// <para>
|
|
/// <list type="bullet">
|
|
/// <item><description>Implement Custom exceptions</description></item>
|
|
/// <item><description>Add unit testing</description></item>
|
|
/// <item><description>Add log4net</description></item>
|
|
/// </list>
|
|
/// </para>
|
|
/// </summary>
|
|
/// <summary>The index of the next character.</summary>
|
|
int myIndex;
|
|
/// <summary>The source string being tokenized.</summary>
|
|
string mySource;
|
|
|
|
# define PROTECTED public
|
|
|
|
# define THROW(x) throw(Error.Generic(x))
|
|
# define SBGET(x) (x)->get()
|
|
|
|
# define int2char(x) String.int2char(x)
|
|
# define trim_whites(x) String.trim_whites(x)
|
|
|
|
program objectbuilder, arraybuilder;
|
|
#else
|
|
# define PROTECTED protected
|
|
|
|
PROTECTED int myIndex;
|
|
PROTECTED string mySource;
|
|
|
|
# define arrayp(x) pointerp(x)
|
|
# define THROW(x) raise_error(x)
|
|
# define UNDEFINED 0
|
|
# define SBGET(x) (x)
|
|
|
|
# define int2char(x) _int2char(x)
|
|
# define trim_whites(x) trim(x)
|
|
# define search strstr
|
|
|
|
PROTECTED mixed nextObject();
|
|
|
|
PROTECTED string _int2char(int c) {
|
|
string s = " ";
|
|
|
|
s[0] = c;
|
|
return s;
|
|
}
|
|
#endif
|
|
|
|
/// <summary>
|
|
/// Construct a JSONTokener from a string.
|
|
/// </summary>
|
|
/// <param name="s">A source string.</param>
|
|
#ifdef __PIKE__
|
|
void create(string s, program|void objectb, program|void arrayb)
|
|
#else
|
|
mixed parse_json(string s)
|
|
#endif
|
|
{
|
|
mySource = s;
|
|
myIndex = 0;
|
|
#ifdef __PIKE__
|
|
objectbuilder = objectb;
|
|
arraybuilder = arrayb;
|
|
#else
|
|
return nextObject();
|
|
#endif
|
|
}
|
|
|
|
/// <summary>
|
|
/// Back up one character. This provides a sort of lookahead capability,
|
|
/// so that you can test for a digit or letter before attempting to parse
|
|
/// the next number or identifier.
|
|
/// </summary>
|
|
PROTECTED void back() {
|
|
if (myIndex > 0)
|
|
myIndex -= 1;
|
|
}
|
|
|
|
#ifndef GAMMA
|
|
// funny.. we already have two better ways to do hex2int than this:
|
|
//
|
|
/// <summary>
|
|
/// Get the hex value of a character (base16).
|
|
/// </summary>
|
|
/// <param name="c">
|
|
/// A character between '0' and '9' or between 'A' and 'F' or
|
|
/// between 'a' and 'f'.
|
|
/// </param>
|
|
/// <returns>An int between 0 and 15, or -1 if c was not a hex digit.</returns>
|
|
PROTECTED int dehexchar(int c) {
|
|
if (c >= '0' && c <= '9')
|
|
{
|
|
return c - '0';
|
|
}
|
|
if (c >= 'A' && c <= 'F')
|
|
{
|
|
return c + 10 - 'A';
|
|
}
|
|
if (c >= 'a' && c <= 'f')
|
|
{
|
|
return c + 10 - 'a';
|
|
}
|
|
return -1;
|
|
}
|
|
#endif
|
|
|
|
/// <summary>
|
|
/// Determine if the source string still contains characters that next() can consume.
|
|
/// </summary>
|
|
/// <returns>true if not yet at the end of the source.</returns>
|
|
PROTECTED int more() {
|
|
return myIndex < sizeof(mySource);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Get the next character in the source string.
|
|
/// </summary>
|
|
/// <returns>The next character, or 0 if past the end of the source string.</returns>
|
|
#ifdef __PIKE__
|
|
PROTECTED int next(int|void x)
|
|
#else
|
|
varargs PROTECTED int next(int x)
|
|
#endif
|
|
{
|
|
if(x) {
|
|
int n = next();
|
|
|
|
if (n != x) {
|
|
THROW("Expected '" + x + "' and instead saw '" + n + "'.\n");
|
|
}
|
|
|
|
return n;
|
|
} else {
|
|
int c = more() ? mySource[myIndex] : 0;
|
|
|
|
myIndex += 1;
|
|
return c;
|
|
}
|
|
}
|
|
|
|
|
|
/// <summary>
|
|
/// Get the next n characters.
|
|
/// </summary>
|
|
/// <param name="n">The number of characters to take.</param>
|
|
/// <returns>A string of n characters.</returns>
|
|
PROTECTED string nextn(int n) {
|
|
int i = myIndex;
|
|
int j = i + n;
|
|
|
|
if (j >= sizeof(mySource)) {
|
|
THROW("Substring bounds error\n");
|
|
}
|
|
myIndex += n;
|
|
return mySource[i..j];
|
|
}
|
|
|
|
/// <summary>
|
|
/// Get the next char in the string, skipping whitespace
|
|
/// and comments (slashslash and slashstar).
|
|
/// </summary>
|
|
/// <returns>A character, or 0 if there are no more characters.</returns>
|
|
PROTECTED int nextClean() {
|
|
while (1) {
|
|
int c = next();
|
|
if (c == '/') {
|
|
switch (next()) {
|
|
case '/':
|
|
do {
|
|
c = next();
|
|
} while (c != '\n' && c != '\r' && c != 0);
|
|
break;
|
|
case '*':
|
|
while (1) {
|
|
c = next();
|
|
|
|
if (c == 0) {
|
|
THROW("Unclosed comment.\n");
|
|
}
|
|
|
|
if (c == '*') {
|
|
if (next() == '/') {
|
|
break;
|
|
}
|
|
|
|
back();
|
|
}
|
|
}
|
|
|
|
break;
|
|
default:
|
|
back();
|
|
return '/';
|
|
}
|
|
}
|
|
else if (c == 0 || c > ' ') {
|
|
return c;
|
|
}
|
|
}
|
|
|
|
#ifndef __PIKE__
|
|
return 0; // will never be reached, stupid LDMud!
|
|
#endif
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return the characters up to the next close quote character.
|
|
/// Backslash processing is done. The formal JSON format does not
|
|
/// allow strings in single quotes, but an implementation is allowed to
|
|
/// accept them.
|
|
/// </summary>
|
|
/// <param name="quote">The quoting character, either " or '</param>
|
|
/// <returns>A String.</returns>
|
|
PROTECTED string nextString(int quote) {
|
|
int c;
|
|
#ifdef __PIKE__
|
|
String.Buffer sb = String.Buffer();
|
|
#else
|
|
string sb = "";
|
|
#endif
|
|
|
|
while (1) {
|
|
c = next();
|
|
if ((c == 0x00) || (c == 0x0A) || (c == 0x0D)) {
|
|
THROW("Unterminated string.\n");
|
|
}
|
|
// CTRL chars
|
|
if (c == '\\') {
|
|
c = next();
|
|
switch (c) {
|
|
case 'b': //Backspace
|
|
sb+="\b";
|
|
break;
|
|
case 't': //Horizontal tab
|
|
sb+="\t";
|
|
break;
|
|
case 'n': //newline
|
|
sb+="\n";
|
|
break;
|
|
case 'f': //Form feed
|
|
sb+="\f";
|
|
break;
|
|
case 'r': // Carriage return
|
|
sb+="\r";
|
|
break;
|
|
case 'u':
|
|
#ifdef __PIKE__
|
|
int iascii;
|
|
sscanf(nextn(4), "%4x", iascii);
|
|
sb+=int2char(iascii);
|
|
#else
|
|
sb+=int2char(hex2int(nextn(2)));
|
|
sb+=int2char(hex2int(nextn(2)));
|
|
#endif
|
|
break;
|
|
default:
|
|
sb+=int2char(c);
|
|
break;
|
|
}
|
|
} else {
|
|
if (c == quote) {
|
|
#ifdef __PIKE__
|
|
return SBGET(sb);
|
|
#else
|
|
return sb;
|
|
#endif
|
|
}
|
|
sb+=int2char(c);
|
|
}
|
|
}//END-while
|
|
|
|
#ifndef __PIKE__
|
|
return ""; // will never be reached. stupid LDMud
|
|
#endif
|
|
}
|
|
|
|
/// <summary>
|
|
/// Unescape the source text. Convert %hh sequences to single characters,
|
|
/// and convert plus to space. There are Web transport systems that insist on
|
|
/// doing unnecessary URL encoding. This provides a way to undo it.
|
|
/// </summary>
|
|
|
|
/// <summary>
|
|
/// Convert %hh sequences to single characters, and convert plus to space.
|
|
/// </summary>
|
|
/// <param name="s">A string that may contain plus and %hh sequences.</param>
|
|
/// <returns>The unescaped string.</returns>
|
|
#ifdef __PIKE__
|
|
PROTECTED string unescape(string|void s)
|
|
#else
|
|
PROTECTED varargs string unescape(string s)
|
|
#endif
|
|
{
|
|
if(!s) s = mySource;
|
|
int len = sizeof(s);
|
|
#ifdef __PIKE__
|
|
String.Buffer sb = String.Buffer();
|
|
#else
|
|
string sb = "";
|
|
#endif
|
|
|
|
for (int i=0; i < len; i++) {
|
|
int c = s[i];
|
|
if (c == '+') {
|
|
c = ' ';
|
|
} else if (c == '%' && (i + 2 < len)) {
|
|
#ifndef GAMMA
|
|
int d = dehexchar(s[i+1]);
|
|
int e = dehexchar(s[i+2]);
|
|
if (d >= 0 && e >= 0) {
|
|
c = (d*16 + e);
|
|
i += 2;
|
|
}
|
|
#else
|
|
i += 2;
|
|
c = hex2int(s[i-1 .. i]);
|
|
#endif
|
|
}
|
|
sb+=int2char(c);
|
|
}
|
|
return SBGET(sb);
|
|
}
|
|
|
|
#ifdef __PIKE__
|
|
mapping|object jsonObject()
|
|
#else
|
|
mapping jsonObject()
|
|
#endif
|
|
{
|
|
#ifdef __PIKE__
|
|
mixed addTo = objectbuilder ? objectbuilder() : ([ ]);
|
|
#else
|
|
mapping addTo = ([ ]);
|
|
#endif
|
|
|
|
if (next() == '%') {
|
|
unescape();
|
|
}
|
|
|
|
back();
|
|
|
|
if (nextClean() != '{') {
|
|
THROW("A JSONObject must begin with '{'.\n");
|
|
}
|
|
|
|
while (1) {
|
|
int c;
|
|
string key;
|
|
mixed obj;
|
|
|
|
c = nextClean();
|
|
switch (c) {
|
|
case 0:
|
|
THROW("A JSONObject must end "
|
|
"with '}'.\n");
|
|
case '}':
|
|
#ifdef __PIKE__
|
|
return mappingp(addTo)
|
|
? addTo
|
|
: ([object]addTo)->finish();
|
|
#else
|
|
return addTo;
|
|
#endif
|
|
case '"':
|
|
back();
|
|
// TODO:: error on != " || '
|
|
key = nextObject();
|
|
break;
|
|
default:
|
|
THROW("Non-String as "
|
|
"JSONObject-key. That "
|
|
"is invalid!\n");
|
|
}
|
|
|
|
if (nextClean() != ':') {
|
|
THROW("Expected a ':' after a key.\n");
|
|
}
|
|
|
|
obj = nextObject();
|
|
|
|
#ifdef __PIKE__
|
|
if (mappingp(addTo)) {
|
|
([mapping]addTo)[key] = obj;
|
|
} else {
|
|
([object]addTo)->add(key, obj);
|
|
}
|
|
#else
|
|
addTo[key] = obj;
|
|
#endif
|
|
|
|
switch (nextClean()) {
|
|
case ',':
|
|
if (nextClean() == '}') {
|
|
#ifdef __PIKE__
|
|
return mappingp(addTo) ? addTo : ([object]addTo)->finish();
|
|
#else
|
|
return addTo;
|
|
#endif
|
|
}
|
|
|
|
back();
|
|
break;
|
|
case '}':
|
|
#ifdef __PIKE__
|
|
return mappingp(addTo) ? addTo : ([object]addTo)->finish();
|
|
#else
|
|
return addTo;
|
|
#endif
|
|
default:
|
|
THROW("Expected a ',' or '}'");
|
|
}
|
|
}
|
|
|
|
#ifdef __PIKE__
|
|
return mappingp(addTo) ? addTo : ([object]addTo)->finish();
|
|
#else
|
|
return addTo;
|
|
#endif
|
|
}
|
|
|
|
#ifdef __PIKE__
|
|
array|object jsonArray()
|
|
#else
|
|
mixed *jsonArray()
|
|
#endif
|
|
{
|
|
#ifdef __PIKE__
|
|
mixed addTo = objectbuilder ? objectbuilder() : ({ });
|
|
#else
|
|
mixed *addTo = ({ });
|
|
#endif
|
|
|
|
if (nextClean() != '[') {
|
|
THROW("A JSONArray must start with '['.\n");
|
|
}
|
|
|
|
if (nextClean() == ']') {
|
|
#ifdef __PIKE__
|
|
return arrayp(addTo) ? addTo : ([object]addTo)->finish();
|
|
#else
|
|
return addTo;
|
|
#endif
|
|
}
|
|
|
|
back();
|
|
while (1) {
|
|
if (arrayp(addTo)) {
|
|
addTo += ({ nextObject() });
|
|
} else {
|
|
addTo->add(nextObject());
|
|
}
|
|
|
|
switch (nextClean())
|
|
{
|
|
case ',':
|
|
if (nextClean() == ']') {
|
|
#ifdef __PIKE__
|
|
return arrayp(addTo)
|
|
? addTo
|
|
: ([object]addTo)->finish();
|
|
#else
|
|
return addTo;
|
|
#endif
|
|
}
|
|
back();
|
|
break;
|
|
case ']':
|
|
#ifdef __PIKE__
|
|
return arrayp(addTo)
|
|
? addTo
|
|
: ([object]addTo)->finish();
|
|
#else
|
|
return addTo;
|
|
#endif
|
|
default:
|
|
THROW("Expected a ',' or ']'.\n");
|
|
}
|
|
}
|
|
|
|
#ifdef __PIKE__
|
|
return arrayp(addTo) ? addTo : ([object]addTo)->finish();
|
|
#else
|
|
return addTo;
|
|
#endif
|
|
}
|
|
|
|
/// <summary>
|
|
/// Get the next value as object. The value can be a Boolean, Double, Integer,
|
|
/// JSONArray, JSONObject, or String, or the JSONObject.NULL object.
|
|
/// </summary>
|
|
/// <returns>An object.</returns>
|
|
PROTECTED mixed nextObject() {
|
|
int c = nextClean();
|
|
string s;
|
|
|
|
if (c == '"' || c == '\'') {
|
|
return nextString(c);
|
|
}
|
|
// Object
|
|
if (c == '{') {
|
|
back();
|
|
return jsonObject();
|
|
}
|
|
|
|
// JSON Array
|
|
if (c == '[') {
|
|
back();
|
|
return jsonArray();
|
|
}
|
|
|
|
#ifdef __PIKE__
|
|
String.Buffer sb = String.Buffer();
|
|
#else
|
|
string sb = "";
|
|
#endif
|
|
|
|
int b = c;
|
|
while (c >= ' ' && c != ':' && c != ',' && c != ']' && c != '}' && c != '/') {
|
|
sb+=int2char(c);
|
|
c = next();
|
|
}
|
|
back();
|
|
|
|
s = trim_whites(SBGET(sb));
|
|
if (s == "true")
|
|
return 1;
|
|
if (s == "false")
|
|
return 0;
|
|
if (s == "null")
|
|
return UNDEFINED;
|
|
|
|
if ((b >= '0' && b <= '9') || b == '.' || b == '-' || b == '+') {
|
|
int a; float b_; string c_;
|
|
sscanf(s, "%d%s", a, c_);
|
|
if(c_ && sizeof(c_)) {
|
|
#ifdef __PIKE__
|
|
sscanf(s, "%f", b_);
|
|
#else
|
|
b_ = to_float(s);
|
|
#endif
|
|
return b_;
|
|
}
|
|
else return a;
|
|
}
|
|
if (s == "") {
|
|
THROW("Missing value.\n");
|
|
}
|
|
return s;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Skip characters until the next character is the requested character.
|
|
/// If the requested character is not found, no characters are skipped.
|
|
/// </summary>
|
|
/// <param name="to">A character to skip to.</param>
|
|
/// <returns>
|
|
/// The requested character, or zero if the requested character is not found.
|
|
/// </returns>
|
|
PROTECTED int skipTo(int to) {
|
|
int c;
|
|
int i = myIndex;
|
|
do {
|
|
c = next();
|
|
if (c == 0) {
|
|
myIndex = i;
|
|
return c;
|
|
}
|
|
}while (c != to);
|
|
|
|
back();
|
|
return c;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Skip characters until past the requested string.
|
|
/// If it is not found, we are left at the end of the source.
|
|
/// </summary>
|
|
/// <param name="to">A string to skip past.</param>
|
|
PROTECTED void skipPast(string to) {
|
|
myIndex = search(mySource, to, myIndex);
|
|
if (myIndex < 0) {
|
|
myIndex = sizeof(mySource);
|
|
} else {
|
|
myIndex += sizeof(to);
|
|
}
|
|
}
|
|
|
|
// TODO implement exception SyntaxError
|
|
|
|
|
|
/// <summary>
|
|
/// Make a printable string of this JSONTokener.
|
|
/// </summary>
|
|
/// <returns>" at character [myIndex] of [mySource]"</returns>
|
|
PROTECTED string ToString() {
|
|
return " at character " + myIndex + " of " + mySource;
|
|
}
|
|
|