Currently, PostgreSQL + Ruby 2.5.1 build has failed.

https://travis-ci.org/rails/rails/jobs/427502723#L2940-L2942
https://travis-ci.org/rails/rails/jobs/427496382#L2921-L2923
https://travis-ci.org/rails/rails/jobs/427470858#L2921-L2923

This is because PostgreSQL server abnormally terminates when ramfs
becomes full.
Ref: https://travis-ci.org/rails/rails/jobs/427960998#L7614

This occurs since d54d0c95.
In d54d0c95, `sudo` is used in preprocessing
for setting MySQL.

In Travis, even if specified `sudo: false`, when using `sudo` in build,
sudo-enabled infrastructure is used.
https://docs.travis-ci.com/user/reference/overview/#virtualisation-environment-vs-operating-system

As container-based infrastructure was changed to sudo-enabled infrastructure,
the disk settings etc were also changed, it seems that this phenomenon occurred.

This can also be handled by extending ramfs(https://github.com/travis-ci/travis-ci/issues/9036#issuecomment-357189213).
However, using sudo-enabled in all builds affects the execution time of
build, so I think that it is good to use sudo-enabled only in the necessary builds.

* Use utf8mb4 character set by default

`utf8mb4` character set supports supplementary characters including emoji.
`utf8` character set with 3-Byte encoding is not enough to support them.

There was a downside of 4-Byte length character set with MySQL 5.5 and 5.6:

"ERROR 1071 (42000): Specified key was too long; max key length is 767 bytes"
for Rails string data type which is mapped to varchar(255) type.

MySQL 5.7 supports 3072 byte key prefix length by default.

* Remove `DEFAULT COLLATE` from Active Record unit test databases

There should be no "one size fits all" collation in MySQL 5.7.
Let MySQL server choose the default collation for Active Record
unit test databases.

Users can choose their best collation for their databases
by setting `options[:collation]` based on their requirements.

* InnoDB FULLTEXT indexes support since MySQL 5.6
it does not have to use MyISAM storage engine whose maximum key length is 1000 bytes.
Using MyISAM storag engine with utf8mb4 character set would cause
"Specified key was too long; max key length is 1000 bytes"

https://dev.mysql.com/doc/refman/5.6/en/innodb-fulltext-index.html

* References

"10.9.1 The utf8mb4 Character Set (4-Byte UTF-8 Unicode Encoding)"
https://dev.mysql.com/doc/refman/5.7/en/charset-unicode-utf8mb4.html

"10.9.2 The utf8mb3 Character Set (3-Byte UTF-8 Unicode Encoding)"
https://dev.mysql.com/doc/refman/5.7/en/charset-unicode-utf8.html

"14.8.1.7 Limits on InnoDB Tables"
https://dev.mysql.com/doc/refman/5.7/en/innodb-restrictions.html
> If innodb_large_prefix is enabled (the default), the index key prefix limit is 3072 bytes
> for InnoDB tables that use DYNAMIC or COMPRESSED row format.

* CI against MySQL 5.7

Followed this instruction and changed root password to empty string.
https://docs.travis-ci.com/user/database-setup/#MySQL-57

* The recommended minimum version of MySQL is 5.7.9
to support utf8mb4 character set and `innodb_default_row_format`

MySQL 5.7.9 introduces `innodb_default_row_format` to support 3072 byte length index by default.
Users do not have to change MySQL database configuration to support Rails string type.

https://dev.mysql.com/doc/refman/5.7/en/innodb-parameters.html#sysvar_innodb_default_row_format

https://dev.mysql.com/doc/refman/5.7/en/innodb-restrictions.html
> If innodb_large_prefix is enabled (the default),
> the index key prefix limit is 3072 bytes for InnoDB tables that use DYNAMIC or COMPRESSED row format.

* The recommended minimum version of MariaDB is 10.2.2
MariaDB 10.2.2 is the first version of MariaDB supporting `innodb_default_row_format`
Also MariaDB says "MySQL 5.7 is compatible with MariaDB 10.2".

- innodb_default_row_format
https://mariadb.com/kb/en/library/xtradbinnodb-server-system-variables/#innodb_default_row_format

- "MariaDB versus MySQL - Compatibility"
https://mariadb.com/kb/en/library/mariadb-vs-mysql-compatibility/
> MySQL 5.7 is compatible with MariaDB 10.2

- "Supported Character Sets and Collations"
https://mariadb.com/kb/en/library/supported-character-sets-and-collations/

This reverts commit 691addbf.

Reason: v10.5.0 has been released, and it seems that the issue is fixed.
Ref: https://github.com/nodejs/node/blob/master/doc/changelogs/CHANGELOG_V10.md#10.5.0

If use Node.js v10.4.0, ujs's build will result in an error, so
temporarily fix the version to v10.3.0.

Ref: https://travis-ci.org/rails/rails/jobs/389960856

- MariaDB 10.3.7 is the first GA release
https://mariadb.com/kb/en/library/mariadb-1037-release-notes/

- MariaDB 10.3 translates `LENGTH()` to `OCTET_LENGTH()` function

https://mariadb.com/kb/en/library/sql_modeoracle-from-mariadb-103/
> MariaDB translates LENGTH() to OCTET_LENGTH()

- MySQL does NOT translate `LENGTH()` to `OCTET_LENGTH()`
However, it translates `OCTET_LENGTH()` to `LENGTH()`

Here are generated schema dumps of this test to show the differences
between MySQL and MariaDB:

* MySQL 8.0 (Server version: 8.0.11 MySQL Community Server - GPL)
```ruby
  create_table \"virtual_columns\", options: \"ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci\", force: :cascade do |t|
    t.string \"name\"
    t.virtual \"upper_name\", type: :string, as: \"upper(`name`)\"
    t.virtual \"name_length\", type: :integer, as: \"length(`name`)\", stored: true
    t.virtual \"name_octet_length\", type: :integer, as: \"length(`name`)\", stored: true
  end
```

* Maria DB 10.3 (Server version: 10.3.7-MariaDB-1:10.3.7+maria~bionic-log mariadb.org binary distribution)

```ruby
  create_table \"virtual_columns\", options: \"ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci\", force: :cascade do |t|
    t.string \"name\"
    t.virtual \"upper_name\", type: :string, as: \"ucase(`name`)\"
    t.virtual \"name_length\", type: :integer, as: \"octet_length(`name`)\", stored: true
    t.virtual \"name_octet_length\", type: :integer, as: \"octet_length(`name`)\", stored: true
  end
```

mutool is licensed under the Affero GPL, which has strict distribution requirements.

Poppler is licensed under the more liberal GPL, making it a good alternative for those who can't use mutool.

Since #32034 Rails 6 requires Ruby 2.4.1 or higher.
Two CI jobs configured with the latest version of`jruby-9.1.15.0`
compatibile with Ruby 2.3.3 are getting errors:

https://travis-ci.org/rails/rails/jobs/343519339

```
Bundler could not find compatible versions for gem "ruby":
  In Gemfile:
    ruby java
    rails java was resolved to 6.0.0.alpha, which depends on
      ruby (>= 2.4.1) java
Could not find gem 'ruby (>= 2.4.1)', which is required by gem 'rails', in any
of the relevant sources:
```

So `2.4` would match `2.4.1` (due to an rvm alias) rather than matching
the latest `2.4.3` release.

Skipping over 2.4.0 to sidestep the `"symbol_from_string".to_sym.dup` bug.

References #32028

Rails 6 will only support Ruby >= 2.3.

I'm not sure cause, but due to the influence of `gem clean`, the expected
gem seems not to be installed correctly.
In order to avoid a test failure due to this, I fixed that `gem clean` not be executed.
Ref: https://github.com/travis-ci/travis-ci/issues/2518#issuecomment-121168856

This is a workaround. If Travis fixes something, please revert this.

Refer https://github.com/rubygems/rubygems/issues/2123

https://www.ruby-lang.org/en/news/2017/12/25/ruby-2-5-0-released/

  RubyGems 2.7 promote bundler-1.16 to default gems.
  It conflicts Travis environment and user installed bundler.

This reverts commit 6ec0ed67.

JRuby 9.1.15.0 has been released:
http://jruby.org/2017/12/07/jruby-9-1-15-0.html

These are needed when GEM=ast, thanks @georgeclaghorn

JRuby 9.1.14.0 has been released and this version is available on Travis CI.
http://jruby.org/2017/11/08/jruby-9-1-14-0

Since 8f2490b5, the integration test of sidekiq and resque is not working
in CI.
https://travis-ci.org/rails/rails/jobs/301276197#L2055
https://travis-ci.org/rails/rails/jobs/301276197#L2061

Because 8f2490b5 removed password from `redis-server`.
So must also remove passwords from these tests.

Simplify our dev testing and CI story since we're also testing against
Redis for the Active Support cache store.

Directly test whether db, host, password, etc are passed through as
config instead of spinning up a Redis server with a password set on it.

This reverts commit 6f481e05, reversing
changes made to 592f790b.

In favor of #31039.

when tested with ruby-head

This pull request attempts to ignore the following error
when tested with ruby-head which has bundler as a default gem.

```ruby
$ rvm @global do gem uninstall bundler --all --ignore-dependencies --executables
ERROR:  While executing gem ... (Gem::InstallError)

    gem "bundler" cannot be uninstalled because it is a default gem

The command "rvm @global do gem uninstall bundler --all --ignore-dependencies --executables" failed and exited with 1 during .
```

Refer https://travis-ci.org/rails/rails/jobs/295600391

This workaround should be removed once https://github.com/bundler/bundler/issues/6072 is addressed.

Since 1.16.0 is installed by default, it seems that the newer one will be used
even if specify an older version.
Ref: https://travis-ci.org/rails/rails/jobs/295553738#L1718

Follow up of #31023

It's causing bug report template tests to fail. https://travis-ci.org/rails/rails/jobs/295520851
This seems an issue of bundler. Ref: https://github.com/bundler/bundler/issues/6072

Most interesting to us, 2.4.2 and 2.3.5 include the fix for
https://bugs.ruby-lang.org/issues/13632

http://jruby.org/2017/09/06/jruby-9-1-13-0.html

Travis CI new default Ubuntu Trusty supports 9.6
https://docs.travis-ci.com/user/database-setup/#Using-a-different-PostgreSQL-Version