mirror of
https://github.com/Ttanasart-pt/Pixel-Composer.git
synced 2024-11-10 20:45:35 +01:00
135 lines
5.3 KiB
Plaintext
135 lines
5.3 KiB
Plaintext
/// json_beautify(json_string)
|
|
// initialization
|
|
// in old versions of GMS, you'd have this ran separately instead.
|
|
// in GMS2 it'd need to be @"..." instead of just "..."
|
|
gml_pragma("global", @"
|
|
global.g_json_beautify_fb = buffer_create(1024, buffer_fast, 1);
|
|
global.g_json_beautify_rb = buffer_create(1024, buffer_grow, 1);
|
|
");
|
|
|
|
function buffer_write_slice(buffer, data_buffer, data_start, data_end) {
|
|
var next = data_end - data_start;
|
|
if (next <= 0) exit;
|
|
var size = buffer_get_size(buffer);
|
|
var pos = buffer_tell(buffer);
|
|
var need = pos + next;
|
|
if (size < need) {
|
|
do size *= 2 until (size >= need);
|
|
buffer_resize(buffer, size);
|
|
}
|
|
buffer_copy(data_buffer, data_start, next, buffer, pos);
|
|
buffer_seek(buffer, buffer_seek_relative, next);
|
|
}
|
|
|
|
function json_beautify(src) {
|
|
// copy text to string buffer:
|
|
var rb = global.g_json_beautify_rb;
|
|
buffer_seek(rb, buffer_seek_start, 0);
|
|
buffer_write(rb, buffer_string, src);
|
|
var size = buffer_tell(rb) - 1;
|
|
var rbsize = buffer_get_size(rb);
|
|
// then copy it to "fast" input buffer for peeking:
|
|
var fb = global.g_json_beautify_fb;
|
|
if (buffer_get_size(fb) < size) buffer_resize(fb, size);
|
|
buffer_copy(rb, 0, size, fb, 0);
|
|
buffer_seek(rb, buffer_seek_start, 0);
|
|
//
|
|
var rbpos = 0; // writing position in output buffer
|
|
var start = 0; // start offset in input buffer
|
|
var pos = 0; // reading position in input buffer
|
|
var next; // number of bytes to be copied
|
|
var need;
|
|
var nest = 0;
|
|
while (pos < size) {
|
|
var c = buffer_peek(fb, pos++, buffer_u8);
|
|
switch (c) {
|
|
case 9: case 10: case 13: case 32: // `\t\n\r `
|
|
buffer_write_slice(rb, fb, start, pos - 1);
|
|
// skip over trailing whitespace:
|
|
while (pos < size) {
|
|
switch (buffer_peek(fb, pos, buffer_u8)) {
|
|
case 9: case 10: case 13: case 32: pos += 1; continue;
|
|
// default -> break
|
|
} break;
|
|
}
|
|
start = pos;
|
|
break;
|
|
case 34: // `"`
|
|
while (pos < size) {
|
|
switch (buffer_peek(fb, pos++, buffer_u8)) {
|
|
case 92: pos++; continue; // `\"`
|
|
case 34: break; // `"` -> break
|
|
default: continue; // else
|
|
} break;
|
|
}
|
|
break;
|
|
case ord("["): case ord("{"):
|
|
buffer_write_slice(rb, fb, start, pos);
|
|
// skip over trailing whitespace:
|
|
while (pos < size) {
|
|
switch (buffer_peek(fb, pos, buffer_u8)) {
|
|
case 9: case 10: case 13: case 32: pos += 1; continue;
|
|
// default -> break
|
|
} break;
|
|
}
|
|
// indent or contract `[]`/`{}`
|
|
c = buffer_peek(fb, pos, buffer_u8);
|
|
switch (c) {
|
|
case ord("]"): case ord("}"): // `[]` or `{}`
|
|
buffer_write(rb, buffer_u8, c);
|
|
pos += 1;
|
|
break;
|
|
default: // `[\r\n\t
|
|
buffer_write(rb, buffer_u16, 2573); // `\r\n`
|
|
repeat (++nest) buffer_write(rb, buffer_u8, 9); // `\t`
|
|
}
|
|
start = pos;
|
|
break;
|
|
case ord("]"): case ord("}"):
|
|
buffer_write_slice(rb, fb, start, pos - 1);
|
|
buffer_write(rb, buffer_u16, 2573); // `\r\n`
|
|
repeat (--nest) buffer_write(rb, buffer_u8, 9); // `\t`
|
|
buffer_write(rb, buffer_u8, c);
|
|
start = pos;
|
|
break;
|
|
case ord(","):
|
|
buffer_write_slice(rb, fb, start, pos);
|
|
buffer_write(rb, buffer_u16, 2573); // `\r\n`
|
|
repeat (nest) buffer_write(rb, buffer_u8, 9); // `\t`
|
|
start = pos;
|
|
break;
|
|
case ord(":"):
|
|
if (buffer_peek(fb, pos, buffer_u8) != ord(" ")) {
|
|
buffer_write_slice(rb, fb, start, pos);
|
|
buffer_write(rb, buffer_u8, ord(" "));
|
|
start = pos;
|
|
} else pos += 1;
|
|
break;
|
|
default:
|
|
if (c >= ord("0") && c <= ord("9")) { // `0`..`9`
|
|
var pre = true; // whether reading pre-dot or not
|
|
var till = pos - 1; // index at which meaningful part of the number ends
|
|
while (pos < size) {
|
|
c = buffer_peek(fb, pos, buffer_u8);
|
|
if (c == ord(".")) {
|
|
pre = false; // whether reading pre-dot or not
|
|
pos += 1; // index at which meaningful part of the number ends
|
|
} else if (c >= ord("0") && c <= ord("9")) {
|
|
// write all pre-dot, and till the last non-zero after dot:
|
|
if (pre || c != ord("0")) till = pos;
|
|
pos += 1;
|
|
} else break;
|
|
}
|
|
if (till < pos) { // flush if number can be shortened
|
|
buffer_write_slice(rb, fb, start, till + 1);
|
|
start = pos;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (start == 0) return src; // source string was unchanged
|
|
buffer_write_slice(rb, fb, start, pos);
|
|
buffer_write(rb, buffer_u8, 0); // terminating byte
|
|
buffer_seek(rb, buffer_seek_start, 0);
|
|
return buffer_read(rb, buffer_string);
|
|
} |