How can I use group_concat on an entire subquery?

Question

...without making unnecessary comparisons

I want to get an md5 hash of a range of rows. Due to bandwidth limitations, I want it to happen server-side.

This w

Answer 1

The solution was simply to omit group by 1 = 1 entirely. I had assumed that group_concat would require that I provide it a group, but it can be used directly on a subquery, like so:

select group_concat(id,col1,col2) from
    (select * from some_table
     where id >= 2 and id < 5
     order by id desc) as some_table;

---

Be aware that null values will need to be cast to something concat-friendly, like so:

insert into some_table (col1, col2)
                values ('a', 1),
                       ('b', 11),
                       ('c', NULL),
                       ('d', 25),
                       ('e', 50);

select group_concat(id, col1, col2) from
    (select id, col1, ifnull(col2, 'NULL') as col2
     from some_table
     where id >= 2 and id < 5
     order by id desc) as some_table;

Output:

+------------------------------+
| group_concat(id, col1, col2) |
+------------------------------+
| 2b11,3cNULL,4d25             |
+------------------------------+

---

Another caveat: mysql has a max-length for group_concat defined by the variable: group_concat_max_len. In order to hash a concatenation of n table rows, I needed to:

1. Hash row so that it is represented in 32 bits, regardless of how many columns it has
2. Ensure that group_concat_max_len > (n * 33) (the extra byte accounts for added commas)
3. Hash the group_concat of the hashed rows.

Ultimately I ended up using the client language to examine the name, number, and nullability of each column, and then build queries like this:

select md5(group_concat(row_fingerprint)) from
    (select concat(id, col1, ifnull(col2, 'null')) as row_fingerprint
     from some_table
     where id >= 2 and id < 5
     order by id desc) as foo;

For more detail, you can poke through my code here (see function: find_diff_intervals).