Function build_prop_table

deps/quickjs/unicode_gen.c:1471–1569

#define PROP_BLOCK_LEN 32

void build_prop_table(FILE *f, int prop_index, bool add_index)
{
    int i, j, n, v, offset, code;
    DynBuf dbuf_s, *dbuf = &dbuf_s;
    DynBuf dbuf1_s, *dbuf1 = &dbuf1_s;
    DynBuf dbuf2_s, *dbuf2 = &dbuf2_s;
    const uint32_t *buf;
    int buf_len, block_end_pos, bit;
    char cname[128];

    dbuf_init(dbuf1);

    for(i = 0; i <= CHARCODE_MAX;) {
        v = get_prop(i, prop_index);
        j = i + 1;
        while (j <= CHARCODE_MAX && get_prop(j, prop_index) == v) {
            j++;
        }
        n = j - i;
        if (j == (CHARCODE_MAX + 1) && v == 0)
            break; /* no need to encode last zero run */
        //printf("%05x: %d %d\n", i, n, v);
        dbuf_put_u32(dbuf1, n - 1);
        i += n;
    }

    dbuf_init(dbuf);
    dbuf_init(dbuf2);
    buf = (uint32_t *)dbuf1->buf;
    buf_len = dbuf1->size / sizeof(buf[0]);

    /* the first value is assumed to be 0 */
    assert(get_prop(0, prop_index) == 0);

    block_end_pos = PROP_BLOCK_LEN;
    i = 0;
    code = 0;
    bit = 0;
    while (i < buf_len) {
        if (add_index && dbuf->size >= block_end_pos && bit == 0) {
            offset = (dbuf->size - block_end_pos);
            /* XXX: offset could be larger in case of runs of small
               lengths. Could add code to change the encoding to
               prevent it at the expense of one byte loss */
            assert(offset <= 7);
            v = code | (offset << 21);
            dbuf_putc(dbuf2, v);
            dbuf_putc(dbuf2, v >> 8);
            dbuf_putc(dbuf2, v >> 16);
            block_end_pos += PROP_BLOCK_LEN;
        }

        v = buf[i];
        code += v + 1;
        bit ^= 1;
        if (v < 8 && (i + 1) < buf_len && buf[i + 1] < 8) {
            code += buf[i + 1] + 1;
            bit ^= 1;