* Fix reading comment only encrypted files
When a encrypted file contains only comments then reading that files raises an error:
NoMethodError: undefined method `deep_symbolize_keys' for false:FalseClass
activesupport/lib/active_support/encrypted_configuration.rb:33:in `config'
test/encrypted_configuration_test.rb:52:in `block in <class:EncryptedConfigurationTest>'
This happens because the previous implementation returned a `{}` fallback for blank YAML strings. But it did not handle YAML strings that are present but still do not contain any _usefull_ YAML - like the file created by `Rails::Generators::EncryptedFileGenerator` which looks like this:
# aws:
# access_key_id: 123
# secret_access_key: 345
* Fix coding style violation
* Add backwardscompatible with Psych versions that were shipped with Ruby <2.5
* Do not rely on railties for Active Support test
* Simplify error handling
* Improve test naming
* Simplify file creation in test
Currently, credentials does not check the format when saving. As a result,
incorrect data as yaml is also saved.
If credentials is used in config files., an error will occur in credential
yaml parsing before edit, and will not be able to edit it.
In order to prevent this, verify the format when saving.
Related: #30851
* WIP: Add credentials using a generic EncryptedConfiguration class
This is sketch code so far.
* Flesh out EncryptedConfiguration and test it
* Better name
* Add command and generator for credentials
* Use the Pathnames
* Extract EncryptedFile from EncryptedConfiguration and add serializers
* Test EncryptedFile
* Extract serializer validation
* Stress the point about losing comments
* Allow encrypted configuration to be read without parsing for display
* Use credentials by default and base them on the master key
* Derive secret_key_base in test/dev, source it from credentials in other envs
And document the usage.
* Document the new credentials setup
* Stop generating the secrets.yml file now that we have credentials
* Document what we should have instead
Still need to make it happen, tho.
* [ci skip] Keep wording to `key base`; prefer defaults.
Usually we say we change defaults, not "spec" out a release.
Can't use backticks in our sdoc generated documentation either.
* Abstract away OpenSSL; prefer MessageEncryptor.
* Spare needless new when raising.
* Encrypted file test shouldn't depend on subclass.
* [ci skip] Some woordings.
* Ditch serializer future coding.
* I said flip it. Flip it good.
* [ci skip] Move require_master_key to the real production.rb.
* Add require_master_key to abort the boot process.
In case the master key is required in a certain environment
we should inspect that the key is there and abort if it isn't.
* Print missing key message and exit immediately.
Spares us a lengthy backtrace and prevents further execution.
I've verified the behavior in a test app, but couldn't figure the
test out as loading the app just exits immediately with:
```
/Users/kasperhansen/Documents/code/rails/activesupport/lib/active_support/testing/isolation.rb:23:in `load': marshal data too short (ArgumentError)
from /Users/kasperhansen/Documents/code/rails/activesupport/lib/active_support/testing/isolation.rb:23:in `run'
from /Users/kasperhansen/.rbenv/versions/2.4.1/lib/ruby/gems/2.4.0/gems/minitest-5.10.2/lib/minitest.rb:830:in `run_one_method'
from /Users/kasperhansen/.rbenv/versions/2.4.1/lib/ruby/gems/2.4.0/gems/minitest-5.10.2/lib/minitest/parallel.rb:32:in `block (2 levels) in start'
```
It's likely we need to capture and prevent the exit somehow.
Kernel.stub(:exit) didn't work. Leaving it for tomorrow.
* Fix require_master_key config test.
Loading the app would trigger the `exit 1` per require_master_key's
semantics, which then aborted the test.
Fork and wait for the child process to finish, then inspect the
exit status.
Also check we aborted because of a missing master key, so something
else didn't just abort the boot.
Much <3 to @tenderlove for the tip.
* Support reading/writing configs via methods.
* Skip needless deep symbolizing.
* Remove save; test config reader elsewhere.
* Move secret_key_base check to when we're reading it.
Otherwise we'll abort too soon since we don't assign the secret_key_base
to secrets anymore.
* Add missing string literal comments; require unneeded yaml require.
* ya ya ya, rubocop.
* Add master_key/credentials after bundle.
Then we can reuse the existing message on `rails new bc4`.
It'll look like:
```
Using web-console 3.5.1 from https://github.com/rails/web-console.git (at master@ce985eb)
Using rails 5.2.0.alpha from source at `/Users/kasperhansen/Documents/code/rails`
Using sass-rails 5.0.6
Bundle complete! 16 Gemfile dependencies, 72 gems now installed.
Use `bundle info [gemname]` to see where a bundled gem is installed.
Adding config/master.key to store the master encryption key: 97070158c44b4675b876373a6bc9d5a0
Save this in a password manager your team can access.
If you lose the key, no one, including you, can access anything encrypted with it.
create config/master.key
```
And that'll be executed even if `--skip-bundle` was passed.
* Ensure test app has secret_key_base.
* Assign secret_key_base to app or omit.
* Merge noise
* Split options for dynamic delegation into its own method and use deep symbols to make it work
* Update error to point to credentials instead
* Appease Rubocop
* Validate secret_key_base when reading it.
Instead of relying on the validation in key_generator move that into
secret_key_base itself.
* Fix generator and secrets test.
Manually add config.read_encrypted_secrets since it's not there by default
anymore.
Move mentions of config/secrets.yml to config/credentials.yml.enc.
* Remove files I have no idea how they got here.
* [ci skip] swap secrets for credentials.
* [ci skip] And now, changelogs are coming.
John Hawthorn