Question

I want to sort a text column, where the numeric components should be sorted as numbers. The sorted result should look like this:

chr1
chr1,chr1
chr1,chr2
chr1,chr10
chr2
chr2,chr1
chr2,chr2
chr2,chr10
chr6
chr6,chr1
chr6_ux9
chr6_ux9,chr1
chr7
chr10
chr10,chr1
chr10,chr2
chr10,chr10
chr21
chr21,chr1
chr21,chr2
chr21,chr10
chrx
chrx,chr1
chrx,chr2
chrx,chr10
chry
chry,chr1
chry,chr2
chry,chr10
chrmt
chrmt,chr1
chrmt,chr2
chrmt,chr10
chr25
chr25,chr1
chr25,chr2
chr25,chr10

The following rules apply:

1. chrx is treated as chr22
2. chry is treated as chr23
3. chrmt is treated as chr24
4. chr6_ux9 is a special case that should come after chr6

I tried different ways but not able to find the perfect solution. Please help me if anyone has an idea.

Answer 1

I think I understand now, what you are looking for. You want the numeric components sorted as numbers, not strings. This should work for you:

SELECT col1
FROM   tbl
ORDER  BY string_to_array(
             replace(replace(replace(replace(replace(replace(
                col1
              , 'chrx',     'chr22')
              , 'chry',     'chr23')
              , 'chrmt',    'chr24')
              , 'chr6_ux9', 'chr6.6')
              , ' chr',     'chr')
              , 'chr',      '')
           , ',')::real[];

Sorts columns as depicted in the questin. The textual component (‘chr’) turns out to be redundant noise. After applying all replacements, I strip the noise and cast to a numeric array which can be used in the ORDER BY clause.

While performing the listed substitutions, the special case for chr6_ux9 forces the use of real[] instead of the simpler and faster int[], because the integer type leaves no room between 6 and 7. You also have one column with a space instead of a comma. I added a substitution for that, too. But that’s probably just a typo. After removing the irrelevant string chr, only comma-separated numbers remain, which can be cast to real[].

BTW, replace() is very fast. I have functions with dozens of replace() operations in a row that still perform fast. (regexp_replace() is much slower.)

---

Alternative answer for sorting individual elements

For a sorted output of all values as strings:

SELECT regexp_split_to_table(replace(replace(replace(
          col1
        ,'chrx', 'chr22')
        ,'chry', 'chr23')
        ,'chrmt', 'chr24')
        , ',') AS col1
FROM   tbl
ORDER  BY 1

chr6_ux9 comes after chr6 automatically in this scenario.