tutor_flake

Custom flake8 rules

Rules

• TUT100 (DEPRECATED, covered by 502) - dataclass class variables require annotations
  • Notes: you can trick this rule by renaming dataclass - and there is the potential for some false positives
• TUT101 - dataclasses.dataclass cannot be renamed in import
  • Notes: this is meant to act against the major case of false negatives for TUT100
• TUT2:
  • TUT200 - asyncio.create_task requires the name parameter
  • TUT201 - asyncio.create_task is either awaited or assigned to a variable
    • See Important here on garbage collection for motivation
  • TUT210 - async function call either await or async for or async with
    • If the body of the function is either pass or immediately errors or immediately returns, the function is excused
    • It is expected that some class functions will fail this rule, but it still has utility
  • TUT220 - no async function in a catch block that will catch a cancelled error or finally that is not one of asyncio.wait_for, asyncio.wait or asyncio.sleep, prevents running forever when task is cancelled
  • TUT230 - any call to cancel is passed a message (that is not a None literal)
    • This is a highly opinionated as it assumes that any cancel call that matches the spec of asyncio.Task.cancel is on a task. To enforce this properly you would use typing, but that is a more aggressive step to take
• TUT300 - no expressions in the main body, unless under name == "main", prevents global side effects
• TUT4:
  • TUT400 - detect strings that were likely meant to be f-strings
    • TODO: we should exclude regex-es
  • TUT410 - detect redundant type annotations
  • TUT411 - detect redundant type annotations for generic assignment
    • Don't do x: Foo[bar] = Foo() ; do x = Foo[bar]()
• TUT5
  • TUT500 (DEPRECATED: covered by 502/503 + mypy) - instance variables set in __init__ cannot overlap with class variables
  • TUT501 - class variables must be defined before all functions
  • TUT502 - class variables must be type annotated
    • with exceptions for Enum and IntEnum
    • It is safe to ignore this rule in the case where the variable is inherited - this is hard to detect in flake though
  • TUT503 - class variables must be annotated as ClassVar or Final, with exceptions for:
    • dataclasses
    • NamedTuple
    • Protocol
    • Enum and IntEnum
    • TypedDict
    • pydantic BaseModel and pydantic-numpy NumpyModel
  • TUT510 - No two argument super within a class
  • TUT511 - A child class must call a parent classes method for __init__
    • Skips if inherited from certain classes: Generic, ABC, Protocol, collections.abc and specific defined subclasses
    • Additional classes to skip are configurable with non-init-classes
  • TUT512 - A child class must call a parent classes method for __post_init__
    • All the same exceptions as 511
  • TUT520 - NotImplemented is only allowed within a dunder method on a class
    • Any other usage is very likely incorrect
  • TUT530 - a constructor must return Self type rather than the class type
    • This encourages good use of constructors that play well with inheritence
    • We detect methods whose return type includes the class type without having any input of the type
• TUT6
  • TUT610 - a function definition allows too many positional arguments (configurable with max-definition-positional-args)
  • TUT620 - a function invocation uses too many positional arguments (configurable with max-invocation-positional-args)
  • TUT630 - a function definition has two consecutive positional arguments with identical typing
    • Positional only or key word only arguments are excluded
    • This is to help prevent arguments where ordering matters being misordered but still passing typing
• TUT7
  • TUT700 - Prevent os.path.<func>() function calls
  • TUT710 - Prevent from os import path
  • TUT720 - Prevent import os.path
• TUT8
  • TUT800 - Prevent time.time
    • Note: programmers often use time.time to extract time deltas in seconds, rather than actually fetching the numbers of seconds since the epoch. This has tricky pitfalls, such as clock sync jumps and lack of monotonicity guarantees.
  • TUT810 - Prevent from time import time (see TUT800)

Future Ideas

• Enforce calling await on async methods
  • I would like to do this - but I don't know how to identify if a method is async
• No addition of string literals (or f-strings)
• Should we also forbid os.walk
• Fixtures must end in _fixt
• Constants must be uppercase
• Empty if blocks should be disallowed
  • Motivated by empty if TYPE_CHECKING

Installation and Configuration

In a pyproject.toml, can install like this:

tutor-flake = {git = "ssh://[email protected]/tutorintelligence/tutor_flake.git", rev = "v0.2.0"}

To configure, in setup.cfg

[flake8:local-plugins]
extension = 
    TUT = tutor_flake.plugin:TutorIntelligenceFlakePlugin