8. Standards and Practices for Coding

The ARMI coding standards are a set of guidelines for helping to create a more consistent and clear code base. Subpart 2.7 402 of NQA-1 states, “Software design verification shall evaluate… the design approach and ensure internal completeness, consistency, clarity and correctness.” While these are required by NQA-1, the idea is that an ARMI developer, who is familiar with these coding standards, should be able to jump from one module to another without changing their coding style.

Tip

The overall theme is: Balance clarity with conciseness.

Just try to be as clear as possible, while using as few words as possible.

Important

Most of the guidelines can be broken, but all deviations need to be justified. It is up to the code reviewers to determine whether the justification was adequate.

Developers and reviewers should consult the standards/guidelines while writing and reviewing code to ensure consistency. Code reviewers should make sure to be familiar with the standards, so that their comments are consistent with other reviewers.

8.1. Code formatting with Black

ARMI uses the Python code formatter black. So while developing code in ARMI it is important to remember to us the black formatter before pushing any code to the repo. All changes pushed to ARMI on github.com will be automatically checked to see if they conform to the black code formatter standards.

The black formatter provides 100% consistency in ARMI for: whitespace, line length, trailing commas, and string formatting.

8.2. Remove commented-out code

If you were testing code and you commented out a block, delete it before sending it in for code review/production. If you want to see the old code later, it will still be in the Git history.

8.3. Avoid hard-coding run parameters

Use the global settings object self.cs for most user-setable parameters that determine the run environment, etc. This will help keep the amount of repeated code down.

Also, do not ever code the following things into the code: user names, passwords, or file paths on your computer. Use environmental variables where possible and user-configurable settings elsewhere. You can also use the armi.ROOT variable (for the active code directory) or armi.RES, and some other useful root-level variables.

8.4. Avoid the global keyword

At all costs, avoid use of the global keyword in your code. Using this keyword can, and usually does, create extremely fragile code that is nigh-impossible to use a debugger on. Especially as part of object-oriented programming, this is extremely lazy design. A careful reader might notice that there are several files in ARMI that are currently using the global keyword. These are all schedule for a refactor to remove the use of global. But, for now, changing the code would cause more annoyance for the ARMI ecosystem userbase than fixing it would. Still, all of those instance in ARMI will be fixed soon.

No new uses of global will make it through the ARMI pull request process.

8.5. Naming conventions

Note

There is a good argument to make that ARMI’s use of camelCase makes the code less readable than if ARMI used snake_case. Unfortunately, making the switch now would affect such a large percentage of the API that it would be more hassle for our user base than it is worth to change.

8.5.1. Use meaningful names

Use descriptive names for variables, functions, methods, classes, and files. This might mean using a longer name like correlationMatrix instead of a shorter one like cm.

8.5.2. General conventions

Here are some general naming guidelines that are always applicable, but particularly applicable to public classes, functions, and methods and their signatures (the signature includes the parameters):

  • Variables that you designate as unused should be prefaced with an underscore (_).

  • Do not use Python reserved keywords as variable names.

  • Try to use names that are pronounceable. (Well-established variable names from equations are exceptable.)

  • Keep names concise and expressive. (An exception is test method names, which may be longer and more descriptive.)

  • Avoid abbreviations and acronyms, unless they are well understood by subject-matter experts (e.g. DB for database, XS for cross-sections, BU for burn up). When using acronyms or abbreviations with camelCase or PascalCase:

    • Use the same case for two-letter acronyms/abbreviations (e.g. diskIO, ioOperation)

    • Use different case for acronyms/abbreviations with more than two characters (e.g. renderHtml(), path)

For consistency, use the following naming conventions:

package names

Python packages, i.e. folders with an __init__.py, shall use camelCase.

module names

Python modules, i.e. python files, shall use camelCase.

Caveat: Test modules are prefixed with test_.

module constants

Module-level “constants” shall be all capitals with an underscore separating words.

function names

Functions shall use camelCase. If the function is only intended to be used within that module, prefix it with a single leading underscore to indicate it is “module protected.”

variable names

Use camelCase. In the odd scenario that the variable is not used (e.g. a method returns a tuple and you only want the first item), prefix it with a single leading underscore to indicate it is “module protected.”

class names

Classes shall use PascalCase. If the class is only intended to be inherited by other classes within the module, prefix the class name with an underscore to indicate it is “module protected.”

class attribute, instance attribute and method names

Use camelCase. If the method is only intended to be used within that module, prefix it with a single leading underscore to indicate it is “class protected.”

8.5.3. Naming quick-reference

Item to be named

Public

Private

package

(folder with an __init__.py)

packageName

N/A

module

(a .py file)

moduleName

N/A

module constant

SPEED_OF_LIGHT_IN_METERS_PER_SECOND

_ONE_OVER_PI

method or function

doSomeAction()

_doSomeAction()

class or instance attribute

assemblies

_assemblies

variable names

linearHeatGenerationRate

_unusedDescription

There are not “private” variables, use this for an unused variable.

8.5.4. Common naming conventions within ARMI

Single character variable names are not usually “clear” or “concise”; however, the following variables are a well-established convention within ARMI and should be used by developers:

  • r when referring to a reactor, and

  • o when referring to a operator

Other names are also consistently used throughout ARMI for specific objects:

  • cs when referring to a :py:class:armi.settings.Settings class; this should not be confused with the .settings attribute of ArmiObject.

  • lib when referring to a cross section library (would have been better as xsLib)

8.6. Prefer shorter methods

A method should have one clear purpose. If you are writing a method that does one thing after the other, break it up into multiple methods and have a primary method call them in order. If your method is longer than 100 lines, see if you can’t break it up. This does a few things:

  1. It makes the code easier to read.

  2. It makes the code chunks more reusable.

  3. It makes the code easier to test.

  4. It makes the code easier to profile, for performance.

8.7. Avoid repeating code

In other words, don’t repeat yourself. (D. R. Y.). Repetitious code is harder to read, and harderd for others to update. If you ever find yourself copying and pasting code, consider pulling the repeated code out into it’s own function, or using a loop.

8.8. Public methods should have docstrings

Always create the proper docstrings for all public functions and public classes.

8.9. Unit tests

All ARMI developers are required to write unit tests.

Important

If you add a new function to the code base, you are required to add unit tests to cover that function.

ARMI uses the pytest library to drive tests, therefore tests need to be runnable from the commandline by python -m pytest armi. Furthermore, for consistency:

  • Each individual unit test should take under 10 seconds, on a modern laptop.

  • All unit tests should be placed into a separate module from production code that is prefixed with test_.

  • All unit tests should be written in object-oriented fashion, inheriting from unittest.TestCase.

  • All test method names should start with test_.

  • All test method names should be descriptive. If the test method is not descriptive enough, add a docstring.

  • Unit tests should have at least one assertion.

8.10. Import statements

Python allows many variations on the import statement, including relative imports, renaming and others. We prefer:

  1. one import per line,

  2. no relative imports

  3. no periods

  4. explicit module/namespace usage

8.10.1. Import ordering

For consistency, import packages in this order:

  1. Python built-in packages

  2. External third-party packages

  3. ARMI modules

Place a single line between each of these groups, for example:

1import os
2import math
3
4import numpy
5from matplotlib import pyplot
6
7from armi import runLog

8.11. Don’t create naked exceptions.

When creating try/except blocks, a naked exception is when the except command is not followed by a specific exception type. Naked exceptions hide a lot of sins, particularly unexpected bugs. This article explains the concept well, as well as a few exceptions to this general rule.

Examples:

Bad

>>> try:
>>>     stuff()
>>> except:
>>>     runLog.warning('Some error occurred in stuff().')

Good (for one exception type)

>>> try:
>>>     stuff()
>>> except AttributeError:
>>>     runLog.warning('Some error occurred in stuff().')

Good (for multiple exception types)

>>> try:
>>>     stuff()
>>> except (ZeroDivisionError, FloatingPointError):
>>>     runLog.warning('Some error occurred in stuff().')

8.12. Data model

Any reactor state information that is created by an Interface should be stored in the ARMI data model. The goal is that given minimal information (i.e. case settings and blueprints) ARMI should be able to load an entire reactor simulation from a given database. If you add state data to your modeling that isn’t stored in the reactor, or add new input files, you will break this paradigm and make everyone’s life just a little bit harder.

8.13. Input files

ARMI developers shall use one of the following well-defined, Python-supported, input file formats.

.json

JSON files are used for a variety of data-object representations. There are some limitations of JSON, in that it does not easily support comments. JSON is also very strict.

.yaml

YAML files are like JSON files but can have comments in them.

8.14. Address the pylint warnings

Our pull request system integrates with the automatic code checker, pylint. Any new code you add must have zero pylint warnings or errors.

8.15. General do’s and don’ts

do not use print

ARMI code should not use the print function; use one of the methods within armi.runLog.

do not use super

In preference to the super function, explicitly call the parent object’s method. For example, in an __init__, use ParentClass.__init__(self, plus, additional, arguments).

do not leave TODO statements in production code

If your TODO statement is important, perhaps it should be a GitHub Issue.