This package serves three purposes:
- It documents how Dart static analysis is used internally at Google,
including best practices for the code we write. The documentation is
this
README.md
. - It contains a corresponding sample
analysis_options.yaml
. - It contains occasional small snippets of Dart code that are used in implementing those best practices.
Most of the recommended lints directly implement the guidelines set out in Effective Dart. In a few cases the lints are stricter than the style guide for the sake of consistency. Details below.
See this Medium article for more background.
Here is how static analysis is used internally at Google:
- By default we disallow checking in code with any errors, warnings, or hints.
- The
TODO
hint is a permanent exception. - Deprecation hints are a permanent exception. Deprecations are handled separately on a case by case basis.
unused_element
,unused_field
andunused_local_variable
are allowed.- When a new SDK version adds new errors, warnings or hints, we either clean up everything before switching SDK version or maintain a whitelist of allowed violations so it can be gradually cleaned up.
- The
- Lints are considered and enabled on a case by case basis. When enabling a lint we first clean up all pre-existing violations. After it's enabled, any attempt to check in a further violation will be blocked.
The currently enabled lints can be found in analysis_options.1.9.0.yaml.
Here are the important places where pedantic
is stricter than Effective Dart:
annotate_overrides
is stricter; Effective Dart says nothing about
@override
.
omit_local_variable_types
is stricter; Effective Dart only says to
avoid
type annotating initialized local variables.
prefer_single_quotes
is stricter; Effective Dart says nothing about
single vs double quotes.
use_function_type_syntax
is stricter; Effective Dart only says to
consider
using the new syntax.
To use the lints add a dependency in your pubspec.yaml
:
# If you use `package:pedantic/pedantic.dart`, add a normal dependency.
dependencies:
pedantic: ^1.9.0
# Or, if you just want `analysis_options.yaml`, it can be a dev dependency.
dev_dependencies:
pedantic: ^1.9.0
then, add an include in your analysis_options.yaml
. If you want to always
use the latest version of the lints, add a dependency on the main
analysis_options
file:
include: package:pedantic/analysis_options.yaml
If your continuous build and/or presubmit check lints then they will likely
fail whenever a new version of package:pedantic
is released. To avoid this,
specify a specific version of analysis_options.yaml
instead:
include: package:pedantic/analysis_options.1.9.0.yaml
The following lints have been considered and will not be enforced:
always_put_control_body_on_new_line
violates Effective Dart "DO format your code using dartfmt". See note about
Flutter SDK style below.
always_put_required_named_parameters_first
does not allow for other logical orderings of parameters, such as matching the
class field order or alphabetical.
always_specify_types
violates Effective Dart "AVOID type annotating initialized local variables"
and others. See note about Flutter SDK style below.
avoid_annotating_with_dynamic
violates Effective Dart "PREFER annotating with dynamic
instead of letting
inference fail".
avoid_as
does not reflect common usage. See note about Flutter SDK style below.
avoid_catches_without_on_clauses
is too strict, catching anything is useful and common even if not always the
most correct thing to do.
avoid_catching_errors
is too strict, the distinction between Error
and Exception
is followed
closely in the SDK but is impractical to follow everywhere.
avoid_classes_with_only_static_members
is too strict, Effective Dart explicitly calls out some cases (constants,
enum-like types) where it makes sense to violate this lint.
avoid_double_and_int_checks
only applies to web, but there is currently no mechanism for enabling a lint
on web code only.
avoid_equals_and_hash_code_on_mutable_classes
would require the @immutable
annotation to be consistently and correctly
used everywhere.
avoid_field_initializers_in_const_classes
does not offer a clear readability benefit.
avoid_js_rounded_ints
only applies to web, but there is currently no mechanism for enabling a lint
on web code only.
avoid_print
is too strict, it's okay to print
in some code.
avoid_returning_null
will be obsoleted by NNBD.
avoid_returning_this
has occasional false positives, and adds little value as the cascade operator
for fluent interfaces in Dart is already very popular.
avoid_slow_async_io
gives wrong advice if the underlying filesystem has unusual properties, for
example a network mount.
avoid_void_async
prevents a valid style where an asynchronous method has a void
return type
to indicate that nobody should await
for the result.
cancel_subscriptions
has false positives when you use a utility method or class to call cancel
.
close_sinks
has false positives when you use a utility method or class to call close
.
constant_identifier_names
is too strict as it does not support the exceptions allowed in Effective Dart
for use of ALL_CAPS.
control_flow_in_finally
does not offer enough value: people are unlikely to do this by accident,
and there are occasional valid uses.
directives_ordering
would enforce a slightly different ordering to that used by IntelliJ and other
tools using the analyzer.
empty_statements
is superfluous, enforcing use of dartfmt
is sufficient to make empty
statements obvious.
flutter_style_todos
is for Flutter SDK internal use, see note about Flutter SDK style below.
invariant_booleans
is experimental.
join_return_with_assignment
does not reflect common usage.
literal_only_boolean_expressions
does not offer enough value: such expressions are easily spotted and so hard
to add by accident.
no_adjacent_strings_in_list
does not offer enough value: adjacent strings in lists are easily spotted
when the code is formatted with dartfmt
.
one_member_abstracts
is too strict, classes might implement more than one such abstract class and
there is no equivalent way to do this using functions.
parameter_assignments
does not reflect common usage, and will cause particular problems with NNBD
code.
prefer_asserts_in_initializer_lists
does not reflect common usage.
prefer_asserts_with_message
is too strict; some asserts do not benefit from documentation.
prefer_bool_in_asserts
is obsolete in Dart 2; bool is required in asserts.
prefer_const_constructors
would add a lot of noise to code now that new
is optional.
prefer_const_constructors_in_immutables
does not often apply as @immutable
is not widely used.
prefer_const_literals_to_create_immutables
is too strict, requiring const
everywhere adds noise.
prefer_constructors_over_static_methods
is too strict, in some cases static methods are better.
prefer_double_quotes
does not reflect common usage.
prefer_expression_function_bodies
is too strict, braces look better for long expressions.
prefer_final_in_for_each
does not reflect common usage.
prefer_final_locals
does not reflect common usage.
prefer_foreach
is too strict; forEach
is not always an improvement.
prefer_int_literals
does not reflect common usage.
prefer_typing_uninitialized_variables
will be obsoleted by NNBD, which comes with type inference for uninitialized
variables.
sort_constructors_first
does not reflect common usage.
sort_unnamed_constructors_first
is too strict, people are free to choose the best constructor ordering.
test_types_in_equals
does not offer enough value: there are plenty of other mistakes possible in
operator ==
implementations. It's better to use codegen.
throw_in_finally
does not offer enough value: people are unlikely to do this by accident,
and there are occasional valid uses.
unnecessary_null_aware_assignments
does not offer enough value: this is hard to do by mistake, and harmless.
use_setters_to_change_properties
is too strict: it can't detect when something is conceptually a property.
use_to_and_as_if_applicable
is too strict: it can't detect when the style rule actually applies.
Note on Flutter SDK Style: some lints were created specifically to support Flutter SDK development. Flutter app developers should instead use standard Dart style as described in Effective Dart, and should not use these lints.
Please file feature requests and bugs at the issue tracker.