.rubocop.yml

# We upgrade to new versions of rubocop manually so it won't harm to enable all
# new cops.
AllCops:
  TargetRubyVersion: 2.5
  NewCops: enable
  SuggestExtensions: false

Style/ArgumentsForwarding:
  Enabled: false

# We don't want to be switching between styles depending on whether
# we're interpolating or not.
Style/StringLiterals:
  EnforcedStyle: double_quotes

Style/SymbolArray:
  EnforcedStyle: brackets

# Block chains allow processing to be broken down into discrete steps
# similarly to Elixir pipelines. They actually increase readability so
# forbidding them isn't a good idea.
Style/MultilineBlockChain:
  Enabled: false

# For consistency across all method definitions.
Style/EmptyMethod:
  EnforcedStyle: expanded

Style/WordArray:
  EnforcedStyle: brackets

Style/IfUnlessModifier:
  Enabled: false
Style/SymbolProc:
  Enabled: false

# Allow explicit nil else clauses.
Style/EmptyCaseCondition:
  Enabled: false

Style/EmptyElse:
  Enabled: false

# if !predicate is often more readable than unless predicate.
Style/NegatedIf:
  Enabled: false

# We allow the old key-value syntax (key => value) in Rakefiles as that's how
# we define task dependencies.
Style/HashSyntax:
  Exclude:
    - lib/active_record_doctor/rake/task.rb
    - '**/*.rake'

# We support older Rubies where those begins are required.
Style/RedundantBegin:
  Enabled: false

# We expect #raise to receive an instance of the exception to raise instead of
# passing in a message as the second parameter. That's because we'd like to be
# able to pass arbitrary parameters to the exception being raised for later
# inspection and processing.
Style/RaiseArgs:
  EnforcedStyle: compact

# There's a tremendous difference between warn and $stderr.puts. We use the
# latter to report some exceptions in a user-friendly manner and don't want to
# mix them up with warnings.
Style/StderrPuts:
  Enabled: false

# We pass an empty block in a few places to have a placeholder to add code to
# in the future without having to dig through the code to determine whether a
# block is allowed. Additionally, removing an empty block is **NOT** a
# refactoring because the method can change it's behavior via #block_given?.
Lint/EmptyBlock:
  Enabled: false

Layout/FirstArrayElementIndentation:
  EnforcedStyle: consistent

Layout/ArgumentAlignment:
  EnforcedStyle: with_fixed_indentation

# <<~ is unsupported on ancient Rubies.
Layout/HeredocIndentation:
  Enabled: false

# Method length is a poor metric for quality as being forced to break a method
# up into multiple smaller pieces can increase cognitive overhead, cause
# confusion (are those smaller methods used anywhere else), add more
# meaningless identifiers.
Metrics/MethodLength:
  Enabled: false

# ABC, cycolmatic complexity and perceived complexity fare rather poorly on our
# detector definitions as trying to satisfy them would mean breaking longish but
# not cognitively demanding method definitions into multiple smaller methods
# which would obscure what's going on by adding lots of indirection.

Metrics/AbcSize:
  Enabled: false

Metrics/CyclomaticComplexity:
  Enabled: false

Metrics/PerceivedComplexity:
  Enabled: false

Metrics/ClassLength:
  Enabled: false

# We know how to decide whether a block is too long or not.
Metrics/BlockLength:
  Enabled: false

# In general, it's a good recommendation but there are two exceptions in
# active_record_doctor where it harms readability. has_presence_validator?
# is more intention revealing than presence_validator? Does the former take
# a model as a parameter and check whether it has a presence validator? Or does
# it take a validator and check its type?
Naming/PredicateName:
  AllowedMethods:
    - "has_presence_validator?"
    - "has_mandatory_presence_validator?"

Gemspec/DevelopmentDependencies:
  EnforcedStyle: gemspec

# We support a broad range of versions and it's impossible to satisfy this.
Gemspec/RequiredRubyVersion:
  Enabled: false

Naming/MethodParameterName:
  AllowedNames:
    - as
    - io

Metrics/ParameterLists:
  CountKeywordArgs: false