Laravel 5.4 upgrade, converting to utf4mb from utf

Question

I'm busy upgrading one of my sites to Laravel 5.4 from 5.3. I noticed when going through the current Github repository that the default character set and collation has been changed from utf8 to utf8mb4 in order to provide support for emojis.

My current database (MariaDB 10.0.29) is currently set to use utf8 but I would however like to upgrade it to use utf8mb4. Unfortunately, I haven't been able to find any documentation regarding this process.

Perhaps I'm overthinking it, but I would have thought that changing the character set and collation of a database would require some work, at very least running some ALTER TABLE commands.

• Is it necessary to go change the character set and collation on the database and tables? Or would it suffice to simply change the settings in my config/database.php file?
• If so, would anyone be able to provide an example migration (Or some MySQL code) on how to achieve this, bearing in mind that preserving existing data is essential.

Thanks

Answer 1

The database charset and collation is the default for newly created tables. The table settings are the defaults for columns.

Do this for each table:

ALTER TABLE table_name CONVERT TO utf8mb4;

Answer 2

Alrighty then, I have written a migration to achieve this for my own system.

• It allows you to optionally specify a connection name to reference a connection other than the default.

• It gets the list of tables from the connection's database using a SHOW TABLES query.

• It then loops through each table and updates all string/character type columns to the new character set and collation.

• I have made it so that a callback must be provided to determine whether or not a column should have it's length changed to the provided new length. In my implementation, VARCHAR and CHAR columns with lengths greater than 191 are updated to have length 191 during the up migration and VARCHAR and CHAR columns with length exactly 191 are updated to have length 255 on the reverse/down migration.

• Once all the string/character columns have been updated, a couple of queries will be ran to change the charset and collation of the table, converting any remaining collations to the new one and then to change the default charset and collation of the table.

• Finally, the database's default charset and collation will be changed.

Notes

• Originally, I tried to simply convert the tables to the new encoding but ran into issues with column lengths. 191 characters is the maximum character length in utf8mb4 when using InnoDB in my version of MySQL/MariaDB and changing the table collation resulted in an error.

• I did at first only want to just update the lengths to the new length but I also wanted to provide a rollback feature, so this was not an option because in the reverse method I would have been setting the lengths of columns that were utf8mb4 to 255, which would have been too long, so I opted to change the collation too.

• I then tried to just change the length, charset and collation of varchar and char columns that were too long, but in my system, this resulted in errors when I had multi-column indexes that included such columns. Apparently, multi-column indexes must use the same collation.

• An important note on this is that the reverse/down migration is not going to be 100% perfect for everyone. I don't think it would be possible to do so without storing extra information about the original columns when migrating. So my current implementation for the reverse/down migration is to assume that columns with length 191 were originally 255.

• A similarly important note on this is that this will blindly change the collations of all string/character columns to the new collation, regardless of the original collation, so if there are columns with different collations, they will all be converted to the new one and the reverse will do the same, the originals won't be preserved.

---

<?php

use Illuminate\Database\Migrations\Migration;

class UpgradeDatabaseToUtf8mb4 extends Migration
{
    /**
     * Run the migrations.
     *
     * @return void
     */
    public function up()
    {
        $this->changeDatabaseCharacterSetAndCollation('utf8mb4', 'utf8mb4_unicode_ci', 191, function ($column) {
            return $this->isStringTypeWithLength($column) && $column['type_brackets'] > 191;
        });
    }

    /**
     * Reverse the migrations.
     *
     * @return void
     */
    public function down()
    {
        $this->changeDatabaseCharacterSetAndCollation('utf8', 'utf8_unicode_ci', 255, function ($column) {
            return $this->isStringTypeWithLength($column) && $column['type_brackets'] == 191;
        });
    }

    /**
     * Change the database referred to by the connection (null is the default connection) to the provided character set
     * (e.g. utf8mb4) and collation (e.g. utf8mb4_unicode_ci). It may be necessary to change the length of some fixed
     * length columns such as char and varchar to work with the new encoding. In which case the new length of such
     * columns and a callback to determine whether or not that particular column should be altered may be provided. If a
     * connection other than the default connection is to be changed, the string referring to the connection may be
     * provided as the last parameter (This string will be passed to DB::connection(...) to retrieve an instance of that
     * connection).
     *
     * @param string       $charset
     * @param string       $collation
     * @param null|int     $newColumnLength
     * @param Closure|null $columnLengthCallback
     * @param string|null  $connection
     */
    protected function changeDatabaseCharacterSetAndCollation($charset, $collation, $newColumnLength = null, $columnLengthCallback = null, $connection = null)
    {
        $tables = $this->getTables($connection);

        foreach ($tables as $table) {
            $this->updateColumnsInTable($table, $charset, $collation, $newColumnLength, $columnLengthCallback, $connection);
            $this->convertTableCharacterSetAndCollation($table, $charset, $collation, $connection);
        }

        $this->alterDatabaseCharacterSetAndCollation($charset, $collation, $connection);
    }

    /**
     * Get an instance of the database connection provided with an optional string referring to the connection. This
     * should be null if referring to the default connection.
     *
     * @param string|null $connection
     *
     * @return \Illuminate\Database\Connection
     */
    protected function getDatabaseConnection($connection = null)
    {
        return DB::connection($connection);
    }

    /**
     * Get a list of tables on the provided connection.
     *
     * @param null $connection
     *
     * @return array
     */
    protected function getTables($connection = null)
    {
        $tables = [];

        $results = $this->getDatabaseConnection($connection)->select('SHOW TABLES');
        foreach ($results as $result) {
            foreach ($result as $key => $value) {
                $tables[] = $value;
                break;
            }
        }

        return $tables;
    }

    /**
     * Given a stdClass representing the column, extract the required information in a more accessible format. The array
     * returned will contain the field name, the type of field (Without the length), the length where applicable (or
     * null), true/false indicating the column allowing null values and the default value.
     *
     * @param stdClass $column
     *
     * @return array
     */
    protected function extractInformationFromColumn($column)
    {
        $type = $column->Type;
        $typeBrackets = null;
        $typeEnd = null;

        if (preg_match('/^([a-z]+)(?:\\(([^\\)]+?)\\))?(.*)/i', $type, $matches)) {
            $type = strtolower(trim($matches[1]));

            if (isset($matches[2])) {
                $typeBrackets = trim($matches[2]);
            }

            if (isset($matches[3])) {
                $typeEnd = trim($matches[3]);
            }
        }

        return [
            'field' => $column->Field,
            'type' => $type,
            'type_brackets' => $typeBrackets,
            'type_end' => $typeEnd,
            'null' => strtolower($column->Null) == 'yes',
            'default' => $column->Default,
            'charset' => is_string($column->Collation) && ($pos = strpos($column->Collation, '_')) !== false ? substr($column->Collation, 0, $pos) : null,
            'collation' => $column->Collation
        ];
    }

    /**
     * Tell if the provided column is a string/character type and needs to have it's charset/collation changed.
     *
     * @param string $column
     *
     * @return bool
     */
    protected function isStringType($column)
    {
        return in_array(strtolower($column['type']), ['char', 'varchar', 'tinytext', 'text', 'mediumtext', 'longtext', 'enum', 'set']);
    }

    /**
     * Tell if the provided column is a string/character type with a length.
     *
     * @param string $column
     *
     * @return bool
     */
    protected function isStringTypeWithLength($column)
    {
        return in_array(strtolower($column['type']), ['char', 'varchar']);
    }

    /**
     * Update all of the string/character columns in the database to be the new collation. Additionally, modify the
     * lengths of those columns that have them to be the newLength provided, when the shouldUpdateLength callback passed
     * returns true.
     *
     * @param string        $table
     * @param string        $charset
     * @param string        $collation
     * @param int|null      $newLength
     * @param Closure|null  $shouldUpdateLength
     * @param string|null   $connection
     */
    protected function updateColumnsInTable($table, $charset, $collation, $newLength = null, Closure $shouldUpdateLength = null, $connection = null)
    {
        $columnsToChange = [];

        foreach ($this->getColumnsFromTable($table, $connection) as $column) {
            $column = $this->extractInformationFromColumn($column);

            if ($this->isStringType($column)) {
                $sql = "CHANGE `%field%` `%field%` %type%%brackets% CHARACTER SET %charset% COLLATE %collation% %null% %default%";
                $search = ['%field%', '%type%', '%brackets%', '%charset%', '%collation%', '%null%', '%default%'];
                $replace = [
                    $column['field'],
                    $column['type'],
                    $column['type_brackets'] ? '(' . $column['type_brackets'] . ')' : '',
                    $charset,
                    $collation,
                    $column['null'] ? 'NULL' : 'NOT NULL',
                    is_null($column['default']) ? ($column['null'] ? 'DEFAULT NULL' : '') : 'DEFAULT \'' . $column['default'] . '\''
                ];

                if ($this->isStringTypeWithLength($column) && $shouldUpdateLength($column) && is_int($newLength) && $newLength > 0) {
                    $replace[2] = '(' . $newLength . ')';
                }

                $columnsToChange[] = trim(str_replace($search, $replace, $sql));
            }
        }

        if (count($columnsToChange) > 0) {
            $query = "ALTER TABLE `{$table}` " . implode(', ', $columnsToChange);

            $this->getDatabaseConnection($connection)->update($query);
        }
    }

    /**
     * Get a list of all the columns for the provided table. Returns an array of stdClass objects.
     *
     * @param string      $table
     * @param string|null $connection
     *
     * @return array
     */
    protected function getColumnsFromTable($table, $connection = null)
    {
        return $this->getDatabaseConnection($connection)->select('SHOW FULL COLUMNS FROM ' . $table);
    }

    /**
     * Convert a table's character set and collation.
     *
     * @param string      $table
     * @param string      $charset
     * @param string      $collation
     * @param string|null $connection
     */
    protected function convertTableCharacterSetAndCollation($table, $charset, $collation, $connection = null)
    {
        $query = "ALTER TABLE {$table} CONVERT TO CHARACTER SET {$charset} COLLATE {$collation}";
        $this->getDatabaseConnection($connection)->update($query);

        $query = "ALTER TABLE {$table} DEFAULT CHARACTER SET {$charset} COLLATE {$collation}";
        $this->getDatabaseConnection($connection)->update($query);
    }

    /**
     * Change the entire database's (The database represented by the connection) character set and collation.
     *
     * # Note: This must be done with the unprepared method, as PDO complains that the ALTER DATABASE command is not yet
     *         supported as a prepared statement.
     *
     * @param string      $charset
     * @param string      $collation
     * @param string|null $connection
     */
    protected function alterDatabaseCharacterSetAndCollation($charset, $collation, $connection = null)
    {
        $database = $this->getDatabaseConnection($connection)->getDatabaseName();

        $query = "ALTER DATABASE {$database} CHARACTER SET {$charset} COLLATE {$collation}";

        $this->getDatabaseConnection($connection)->unprepared($query);
    }
}

Please, please, please back up your database before running this. Use at your own risk!