Now the real work:

my $select = $db->prepare('SELECT line_n, word, word_position FROM tex
+t');
$select->execute;
my $text;
while (my ($line, $word, $pos) = $select->fetchrow_array) {
    $text->{$line}[$pos - 1] = $word;
}
print Dumper($text);
[download]

($q=q:Sq=~/;[c](.)(.)/;chr(-||-|5+lengthSq)`"S|oS2"`map{chr |+ord
}map{substrSq`S_+|`|}3E|-|`7**2-3:)=~y+S|`+$1,++print+eval$q,q,a,
[download]

Thank you very much.

Just a question. Since I'm pretty new to Perl I cannot understand that "strange construct" in the while loop:

$text->{$line}[$pos - 1] = $word;

Please, can you give some more details on it?

Then here come the second question/problem. As far I can see keys are the original line numbers and values are references to arrays containing the words of each line. In that case, if I'm not in wrong, keys/lines do not respect the original order of my text. Is it true?

A very general doorway to "strange constructs": Perl Data Structures Cookbook; see in particular HASHES OF ARRAYS.

---

Give a man a fish:  <%-{-{-{-<

  $text->{$line}[$pos - 1] = $word;
       ^ ^      ^
       | |      |
       | hash   |
dereference    array
[download]

I used $pos - 1 as positions start from 1, but array indices start from 0.

Note that the bracket types are the same as when creating the data:

$VAR1 = {
          'c3' => [
                    'Phasellus',
                    'non',
                    'erat'
                  ],
          'b2' => [
                    'consectetuer',
                    'adipiscing',
                    'elit.'
                  ],
          'a1' => [
                    'Lorem',
                    'ipsum',
                    'dolor',
                    'sit',
                    'amet,'
                  ]
        };
[download]

($q=q:Sq=~/;[c](.)(.)/;chr(-||-|5+lengthSq)`"S|oS2"`map{chr |+ord
}map{substrSq`S_+|`|}3E|-|`7**2-3:)=~y+S|`+$1,++print+eval$q,q,a,
[download]

without GROUP_CONCT and similar database functions)

Do you mean without a special group concatenation function(), or without any functions at all? This can be easily done with a recursive query.

WITH
    Builder(line_n, word, word_position, amount)
AS
    (
     SELECT
        line_n,
        '',
        0,
        MAX(word_position)
     FROM
        table
     GROUP BY
        line_n
     UNION ALL
     SELECT
        Data.line_n,
        CONCAT(Builder.word, CONCAT(' ', Data.word)),
        Data.word_position,
        Builder.amount
     FROM
        Builder,
        table
     WHERE
        table.line_n        = table.line_n
       AND    table.word_position    = table.word_position + 1
    )
SELECT
    line_n,
    word
FROM
    Builder
WHERE
    word_position = amount;
[download]

The double CONCAT() was used to avoid casting issues; you may be able to remove them. Depending on your RDBMS, you may need to specify the RECURSIVE keyword.

Also, if you don't mind, a side point, please do not use * outside of COUNT(), EXISTS(), and ad hoc queries. Specifying the columns--id, line_n, word, word_position--is easy, self-documenting, and protects against column changes and reordering.

Perhaps it is worth mentioning, the id column is redundant, as line_n and word_position must be unique anyway.

Thank you for you help, but in fact I would like to avoid SQL functions at all. For instance, GROUP_CONCAT or string_agg would be great. I need to manage these data before their way to the HTML template. Moreover, just to clarify my question, neither line_n nor word_position are unique values.

If line_n and word_position (together) are not unique, how in the world do you build the string? Well, unless id breaks the ties.

If you want to see the work in progress, the recursive query is even easier. (No MAX in the anchor, and no WHERE clause is the main query.)