We generally follow simple plain-English commit titles and summaries, which encapsulate what a commit has done. We don't use a distinct commit style like that of the Linux Kernel for any commits and they would look extremely out of place in the repository overall.
In addition, we stick to a single objective with one commit albeit ensure that the scope isn't too small so there'll be a huge amount of them or too large so it's a single commit that changes vast swaths of the codebase. Try to find the right balance between committing too less and too much.
### Use line-wrapping
There is no column limit in the codebase, this is so that the line width can adjust to everyone's display size using line-wrap. Do not manually wrap lines unless it can be done in a natural way and is needed at all, let line-wrap handle it for you.
### Use code formatter
Android Studio comes with a code formatter in-built, this can fix minor mistakes in code-style. To reformat code: Right-click on the relevant file/directory -> Reformat. We recommend doing so prior to all commits to ensure the codebase is clean.
This can also be done by using `Ctrl + Alt + L` on Windows, `Ctrl + Shift + Alt + L` on Linux and `Option + Command + L` on macOS.
Any code that's been PR'd to the Skyline repository will only be in Edge builds for two weeks, after which it will be merged into the mainline branch. This is to ensure that any bugs that may have been introduced by the PR are caught and fixed before it's merged into the mainline branch. If you have any issues with this, you can request that we add the `CI` tag to your PRs so that CI builds are provided pre-merge.
* Classes/Structs - Use `/**` block comments with a brief
* Class/Struct Variables - Use `//!<` single-line comments on their utility
* Class/Struct Functions - Use `/**` block comments on their function with a brief, all arguments and the return value (The brief can be skipped if the function's arguments and return value alone explain what the function does)
* Enumerations - Use a `/**` block comment with a brief for the enum itself and a `//!<` single-line comment for all the individual items
* The `DeviceState` object or other objects which are systematically used throughout multiple classes such as `ServiceManager` can be skipped from function argument documentation as well as class members in the constructor
* The `KSession`, `IpcRequest` and `IpcResponse` objects in IPC command function arguments and other such objects (Such as `IoctlData`) can be skipped from function argument documentation if used in those contexts, they will need to be documented if they're used as a class member or something on those lines
We generally follow the rule of **"Functional Spacing"**, that being spacing between chunks of code that do something functionally different while functionally similar blocks of code can be closer together.
Here are a few examples of this to help with intution:
We generally support the usage of functional programming and lambda, usage of it for assigning conditional variables is recommended especially if it would otherwise be a nested ternary statement:
For passing any parameter which isn't a primitive prefer to use references/const references to pass them into functions or other places as a copy can be avoided that way.
In addition, always use a const reference rather than a normal reference unless the argument needs to be modified in-place as the compiler knows the intent far better in that case.
Note: In constructors if you are copying to a member variable `std::move` is preferred as it allows copy-elision in some circumstances.
Use C++ range-based iterators for any C++ container iteration unless it can be performed better with functional programming (After C++20 when they are merged with the container). In addition, stick to using references/const references using them
Use `auto` to assign a variable the type of the value it's being assigned to, but not where a different type is desired. So, as a rule of thumb always specify the type when setting something from a number rather than depending on `auto`. In addition, prefer not to use `auto` in cases where it's hard to determine the return type due to assigned value being complex.
We generally use `in` and `un` where `n = {8, 16, 32, 64}` for our integer primitives in which `i` represents signed integers and `u` represents unsigned integers. In addition, we have some other types such as `KHandle` that are used to make certain operations more clear, use these depending on the context.
If a variable is constant at compile time use `constexpr`, if it's only used in a local function then place it in the function but if it's used throughout a class then in the corresponding header add the variable to the `skyline::constant` namespace. If a constant is used throughout the codebase, add it to `common.h`.
In addition, try to `constexpr` as much as possible including constructors and functions so that they may be initialized at compile-time and have lesser runtime overhead during usage and certain values can be pre-calculated in advance.
We should also mention that this isn't promoting the usage of `const`, it's use is actually discouraged out of references, in which case it is extremely encouraged. In addition, pointers are a general exception to this, using `const` with them isn't encouraged nor discouraged. Another exception are class functions, they can be made `const` if used from a `const` reference/pointer and don't
modify any members but do not do this preemptively.
We use bracketed initialization as opposed to traditional initalization due to the better type checking it offers and the consistency with designated initalizers.
* Correct
```c++
int a{FindA()};
static constexpr size_t AConstant{1ULL <<63}
for (int i{}; i <a;i++);
```
* Incorrect
```c++
int a = FindA();
static constexpr size_t AConstant = 1ULL <<63;
for (int i = 0; i <a;i++);
```
### Wrapping
We do not enforce a particular limit on line lengths however excessively long lines that may be difficult to read when soft-wrapped should be wrapped semantically. See the below examples:
The size of the header imported for [Vulkan-Hpp](https://github.com/KhronosGroup/Vulkan-Hpp) is extremely large and exceeds the CLion default analysis limit, it is required to run for properly annotating any code which uses components from it. To override this limit, refer to this [article from JetBrains](https://www.jetbrains.com/help/objc/configuring-file-size-limit.html#file-size-limit) or navigate to Help -> Edit Custom Properties and add `idea.max.intellisense.filesize=20000` to set the maximum limit to 20MB which should be adequate for it.
**(1)** Except when the whole name is an abbreviation such as `OS` and `NCE` but not `JVMManager`
### Comments
Use KDoc comments (`/**`) for:
* Classes - A comment about the function of a class and any of it's parameters if required, albeit this can be skipped if the information is mundane enough to the point of it being useless which is generally the case with [Activities](https://developer.android.com/reference/android/app/Activity), albeit this isn't always the case. In addition, document any parameters in the primary constructor using `@param`.
* Variables/Functions - A comment about what the variable is used for or what it is depending on what's more applicable. This is mandatory and shouldn't be skipped.
* Functions - A comment about what the function is used for or what it is depending on what's more applicable and is mandatory, just like a variable. All parameters should be documented using `@param`. In addition, if it's an overridden function then it's description should cover more what it specifically does rather than what it is.
* Constructors - A comment about the constructor can be skipped if it simply calls another constructor or just assigns variables, assuming those variables are documented themselves. If the constructor does anything outside of that, it must be documented in accordance to the function comments.
Exceptions to `@param`:
* If a parameter has already been tagged in the comment using a KDoc link (`[]`), then it doesn't need to have it's own `@param`
* If a parameter is a standard Android class such as `Context` then it's documentation can be skipped entirely
Notes:
* Any class members don't need to be redundantly documented in the constructor
* Use a KDoc link (`[]`) when referencing any item
### Spacing
The spacing in Kotlin follows the same rule as C++ with "**Functional Spacing**", spacing between chunks of code that do something functionally different while functionally similar blocks of code can be closer together.
Here are a few examples of this to help with intution:
* Spacing should generally follow multiple related variable declarations
```kotlin
val a = GetSomethingA()
val b = GetSomethingB()
val c = GetSomethingC()
val result = DoSomething(a, b, c)
```
* If a function doesn't require multiple variable declarations, the function call should be right after the variable declaration
```kotlin
val a = GetClassA()
a.DoSomething()
val b = GetClassB()
b.DoSomething()
```
```kotlin
val a = GetClassA()
DoSomething(a)
val b = GetClassB()
DoSomething(b)
```
* If a single variable is used by a single-line control flow statement, there can be no spaces after it's declaration
```kotlin
val a = GetClass()
if (a.fail)
throw exception()
```
* In `when` statements, if cases generally have multi-line code blocks then have spacing between all the cases
```kotlin
when(num) {
1 -> {
a.Something()
a.SomethingElse()
}
2 -> {
b.Something()
b.SomethingElse()
}
else -> c.Something()
}
```
```kotlin
when(num) {
1 -> a.Something()
2 -> b.Something()
else -> c.Something()
}
```
* **Be consistent** (The above examples assume that `a`/`b`/`c` are correlated)
* Inconsistent
```kotlin
val a = GetClassA()
a.DoSomething()
val b = GetClassB()
val c = GetClassC()
b.DoSomething()
c.DoSomething()
```
* Consistent #1
```kotlin
val a = GetClassA()
val b = GetClassB()
val c = GetClassC()
a.DoSomething()
b.DoSomething()
c.DoSomething()
```
* Consistent #2
```kotlin
val a = GetClassA()
a.DoSomething()
val b = GetClassB()
b.DoSomething()
val c = GetClassC()
c.DoSomething()
```
### Control flow statements (if, for and while)
#### If a child control-flow statement has brackets, the parent statement must as well
* Correct
```cpp
if (a) {
while (b) {
Something(a)
Something(b)
}
}
```
* Incorrect
```cpp
if (a)
while (b) {
Something(a)
Something(b)
}
```
#### If any cases of an if statement have curly brackets all must have curly brackets