Permalink
228 lines (157 sloc) 11.7 KB

C# Coding Guidelines

Coding Conventions

As a general rule, our coding convention is to follow the style of the surrounding code. So if a file happens to differ in style from conventions defined here (e.g. private members are named m_member rather than _member), the existing style in that file takes precedence.

When making changes, you may find some existing code goes against the conventions defined here. In such cases, please avoid reformatting any existing code when submitting a PR as it obscures the functional changes of the PR. A separate PR should be submitted for style-only changes. We also run the .NET code formatter tool regularly to keep consistent formatting.

Naming Conventions

  • Use meaningful, descriptive words for names. For method names, it's encouraged to use VerbObject pair such as LoadModule.

  • Use _camelCase to name internal and private fields and use readonly where possible. Prefix instance fields with _, static fields with s_ and thread static fields with t_. When used on static fields, readonly should come after static (i.e. static readonly not readonly static).

  • Use camelCase to name non-constant local variables.

  • Use PascalCase to name constant local variables and fields. The only exception is for interop code where the constant should exactly match the name and value of the code you are calling via interop (e.g. const int ERROR_SUCCESS = 0).

  • Use PascalCase to name types and all other type members.

Layout Conventions

  • Use four spaces of indentation (no tabs).

  • Avoid more than one blank empty line at any time.

  • Avoid trailing spaces at the end of a line.

  • Braces usually go on their own lines, with the exception of single line statements that are properly indented.

  • Namespace imports should be specified at the top of the file, outside of namespace declarations.

  • Fields should be specified at the top within type declarations. For those that serve as backing fields for properties, they should be specified next to the corresponding properties.

  • Preprocessor directives like #if and #endif should be placed at the beginning of a line, without any leading spaces.

  • File encoding should be ASCII. All BOM encodings should be avoided. Tests that need a BOM encoding file should generate the file on the fly.

Member Conventions

  • Use of this is neither encouraged nor discouraged.

  • Use nameof(<member-name>) instead of "<member-name>" whenever possible and relevant. The motivation is to easily and more accurately find references.

  • Always specify the visibility, even if it's the default (i.e. private string _foo not string _foo). Visibility should be the first modifier (i.e. public abstract not abstract public).

  • Make members private where possible. Avoid declaring public members unless it's absolutely necessary.

  • Public members in a namespace that ends with Internal, for example System.Management.Automation.Internal are not considered a supported public API. Such members are necessarily public as implementation details in code shared between C# and PowerShell script, or must be available publicly by generated code.

Commenting Conventions

  • Place the comment on a separate line, not at the end of a line of code.

  • Begin comment text with an uppercase letter. It's recommended to end comment text with a period but not required.

  • Add comments where the code is not trivial or could be confusing.

  • Add comments where a reviewer needs help to understand the code.

  • Update/remove existing comments when you are changing the corresponding code.

  • Make sure the added/updated comments are meaningful, accurate and easy to understand.

  • Public members must use doc comments. Internal and private members may use doc comments but it is not required.

Performance Considerations

PowerShell has a lot of performance sensitive code as well as a lot of inefficient code. We have some guidelines that we typically apply widely even in less important code because code and patterns are copied, and we want certain inefficient code to stay out of the performance critical code.

Some general guidelines:

  • Avoid LINQ - it can create lots of avoidable garbage. Instead, iterate through a collection directly using for or foreach loop.

  • Between for and foreach, for is slightly preferred when you're uncertain if foreach allocates an iterator.

  • Avoid params arrays, prefer adding overloads with 1, 2, 3, and maybe more parameters.

  • Be aware of APIs such as String.Split(params char[]) that do not provide overloads to avoid array allocation. When calling such APIs, reuse a static array when possible (e.g. Utils.Separators.Colon).

  • Avoid creating empty arrays. Instead, reuse the static ones via Utils.EmptyArray<T>.

  • Avoid unnecessary memory allocation in a loop. Move the memory allocation outside the loop if possible.

  • Avoid gratuitous exceptions as much as possible. Exception handling can be expensive due to cache misses and page faults when accessing the handling code and data. Finding and designing away exception-heavy code can result in a decent performance win. For example, you should stay away from things like using exceptions for control flow.

  • Avoid if (obj is Example) { example = (Example)obj } when casting an object to a type. Instead, use var example = obj as Example or the C# 7 syntax if (obj is Example example) {...} as appropriate. In this way you can avoid converting to the type twice.

  • Use generic collections instead of the non-generic ones such as ArrayList and Hashtable to avoid type casting and unnecessary boxing whenever possible.

  • Use collection constructor overloads that take an initial capacity for collection types that have them. Internally, List<T>, Dictionary<TKey, TValue>, and the other generic collections use one or more arrays to hold valid data. Whenever resizing is needed, one or more new arrays double the size of existing arrays are created and items from the existing arrays are copied. Setting an approximate initial capacity will reduce the number of resizing operations.

  • Use dict.TryGetValue instead of dict.Contains and dict[..] when retrieving value from a Dictionary. In this way you can avoid hashing the key twice.

  • It's OK to use the + operator to concatenate one-off short strings. But when dealing with strings in loops or large amounts of text, use a StringBuilder object.

Security Considerations

Security is an important aspect of PowerShell and we need to be very careful about changes that may introduce security risks, such as code injection caused by the lack of input validation, privilege escalation due to the misuse of impersonation, or data privacy breach with a plain text password.

Reviewers of a PR should be sensitive to changes that may affect security. Some security related keywords may serve as good indicators, such as password, crypto, encryption, decryption, certificate, authenticate, ssl/tls and protected data.

When facing a PR with such changes, the reviewers should request a designated security Subject Matter Expert (SME) to review the PR. Currently, @PaulHigin and @TravisEz13 are our security SMEs. See CODEOWNERS for more information about the area experts.

Best Practices

  • Avoid hard-coding anything unless it's absolutely necessary.

  • Avoid a method that is too long and complex. In such case, separate it to multiple methods or even a nested class as you see fit.

  • Use using statement instead of try/finally if the only code in the finally block is to call the Dispose method.

  • Use of object initializers (e.g. new Example { Name = "Name", ID = 1 }) is encouraged for better readability, but not required.

  • Stick to the DRY principle -- Don't Repeat Yourself.

    • Wrap the commonly used code in methods, or even put it in a utility class if that makes sense, so that the same code can be reused (e.g. StringToBase64Converter.Base64ToString(string)).
    • Check if the code for the same purpose already exists in the code base before inventing your own wheel.
    • Avoid repeating literal strings in code. Instead, use const variable to hold the string.
    • Resource strings used for errors or UI should be put in resource files (.resx) so that they can be localized later.
  • Use of new C# language syntax is encouraged. But avoid refactoring any existing code using new language syntax when submitting a PR as it obscures the functional changes of the PR. A separate PR should be submitted for such refactoring without any functional changes.

  • Consider using the Interlocked class instead of the lock statement to atomically change simple states. The Interlocked class provides better performance for updates that must be atomic.

  • Here are some useful links for your reference:

Portable Code

There are 3 primary preprocessor macros we use during builds:

  • DEBUG - guard code that should not be included in release builds
  • CORECLR - guard code that differs between Full CLR and CoreCLR
  • UNIX - guard code that is specific to Unix (Linux and macOS)

Any other preprocessor defines found in the source are used for one-off custom builds, typically to help debug specific scenarios.

Here are some general guidelines for writing portable code:

  • We are in the process of cleaning up Full CLR specific code (code enclosed in !CORECLR), so do not use CORECLR or !CORECLR in new code. PowerShell Core targets .NET Core only and all new changes should support .NET Core only.

  • The PowerShell code base started on Windows and depends on many Win32 APIs through P/Invoke. Going forward, we try to depend on .NET Core to handle platform differences, so avoid adding new P/Invoke calls where a suitable alternative exists in .NET Core.

  • Try to minimize the use of #if UNIX. When absolutely necessary, avoid duplicating more code than necessary, and instead prefer introducing helper functions to minimize the platform differences.

  • When adding platform dependent code (Windows vs. UNIX), prefer preprocessor directives over runtime checks. However, runtime checks are acceptable if it would greatly improve readability without causing performance concerns in performance-sensitive code.

  • We produce a single binary for all UNIX variants, so runtime checks are currently necessary for some of them (e.g. macOS vs. Linux).