I think the overhead of str_len is acceptable.

The comparison is not very suitable here. If the command select 000 is issued, OK should be returned.

Yes, are you suggesting pos_to_long?

The second argument of SELECT is the index of database. The name db_data is not quite suitable here.

Well it's the index of db, I think db tells more than index here.

Redis uses strtoll which is available in C99 for error handling. Both of us just missed it!

To make use of strtoll, a pos_array must be converted into a string first, too expensive for an operation like this.

If we try our best to conform to redis, we should consider this:

> select hello
(error) ERR invalid DB index

Agree, I've thought about that.

The error messages are completely different in some cases:

Redis:
> select
(error) ERR wrong number of arguments for 'select' command

Corvus:
> select
(error) ERR Proxy fail to forward command

I'm afraid it's too much work to get them exactly the same in this PR.

And if nothing goes wrong, do we really have to do this?

I think supporting "SELECT is not allowed in cluster mode" is enough.

Yeah it's enough.

length here should be pos->pos_len. See cmd_get_map_key for how to iterate pos_array.

But pos_to_str uses pos->str_len instead of pos->pos_len.

pos_array is used to represent only one string, consisting of split sub string stored in items

struct pos_array {
    struct pos *items;
    int str_len;                    // str_len == sum(items.len)
    int pos_len;                  // pos_len == len(items)
    int max_pos_size;       // max_pos_size is the actual memory size allocated
};

Actually that's what I'm thinking about at the beginning, then I find this and missread .pos_len = 2 into .pos_len = 12.

Thanks for the hint, what about now?