For a lot of people a custom keyboard is about more than sending button presses to your computer. You want to be able to do things that are more complex than simple button presses and macros. QMK has hooks that allow you to inject code, override functionality, and otherwise customize how your keyboard behaves in different situations.
This page does not assume any special knowledge about QMK, but reading [Understanding QMK](understanding_qmk.md) will help you understand what is going on at a more fundamental level.
Each of the functions described below can be defined with a `_kb()` suffix or a `_user()` suffix. We intend for you to use the `_kb()` suffix at the Keyboard/Revision level, while the `_user()` suffix should be used at the Keymap level.
When defining functions at the Keyboard/Revision level it is important that your `_kb()` implementation call `_user()` before executing anything else- otherwise the keymap level function will never be called.
By far the most common task is to change the behavior of an existing keycode or to create a new keycode. From a code standpoint the mechanism for each is very similar.
The first step to creating your own custom keycode(s) is to enumerate them. This means both naming them and assigning a unique number to that keycode. Rather than limit custom keycodes to a fixed range of numbers QMK provides the `SAFE_RANGE` macro. You can use `SAFE_RANGE` when enumerating your custom keycodes to guarantee that you get a unique number.
When you want to override the behavior of an existing key, or define the behavior for a new key, you should use the `process_record_kb()` and `process_record_user()` functions. These are called by QMK during key processing before the actual key event is handled. If these functions return `true` QMK will process the keycodes as usual. That can be handy for extending the functionality of a key rather than replacing it. If these functions return `false` QMK will skip the normal key handling, and it will be up to you to send any key up or down events that are required.
This example does two things. It defines the behavior for a custom keycode called `FOO`, and it supplements our Enter key by playing a tone whenever it is pressed.
*`keyboard_pre_init_*` - Happens before most anything is started. Good for hardware setup that you want running very early.
*`matrix_init_*` - Happens midway through the firmware's startup process. Hardware is initialized, but features may not be yet.
*`keyboard_post_init_*` - Happens at the end of the firmware's startup process. This is where you'd want to put "customization" code, for the most part.
!> For most people, the `keyboard_post_init_user` function is what you want to call. For instance, this is where you want to set up things for RGB Underglow.
* This needs to perform the low-level initialisation of all row and column pins. By default this will initialise the input/output state of each of the GPIO pins listed in `MATRIX_ROW_PINS` and `MATRIX_COL_PINS`, based on whether or not the keyboard is set up for `ROW2COL`, `COL2ROW`, or `DIRECT_PINS`. Should the keyboard designer override this function, no initialisation of pin state will occur within QMK itself, instead deferring to the keyboard's override.
* These three functions need to perform the low-level retrieval of matrix state of relevant input pins, based on the matrix type. Only one of the functions should be implemented, if needed. By default this will iterate through `MATRIX_ROW_PINS` and `MATRIX_COL_PINS`, configuring the inputs and outputs based on whether or not the keyboard is set up for `ROW2COL`, `COL2ROW`, or `DIRECT_PINS`. Should the keyboard designer override this function, no manipulation of matrix GPIO pin state will occur within QMK itself, instead deferring to the keyboard's override.
This is ran as the very last task in the keyboard initialization process. This is useful if you want to make changes to certain features, as they should be initialized by this point.
### Example `keyboard_post_init_user()` Implementation
This example, running after everything else has initialized, sets up the rgb underglow configuration.
```c
void keyboard_post_init_user(void) {
// Call the post init code.
rgblight_enable_noeeprom(); // enables Rgb, without saving settings
Whenever possible you should customize your keyboard by using `process_record_*()` and hooking into events that way, to ensure that your code does not have a negative performance impact on your keyboard. However, in rare cases it is necessary to hook into the matrix scanning. Be extremely careful with the performance of code in these functions, as it will be called at least 10 times per second.
This example has been deliberately omitted. You should understand enough about QMK internals to write this without an example before hooking into such a performance sensitive area. If you need help please [open an issue](https://github.com/qmk/qmk_firmware/issues/new) or [chat with us on Discord](https://discord.gg/Uq7gcHh).
This function gets called at every matrix scan, which is basically as often as the MCU can handle. Be careful what you put here, as it will get run a lot.
You should use this function if you need custom matrix scanning code. It can also be used for custom status output (such as LEDs or a display) or other functionality that you want to trigger regularly even when the user isn't typing.
This function gets called at the end of all QMK processing, before starting the next iteration. You can safely assume that QMK has dealt with the last matrix scan at the time that these functions are invoked -- layer states have been updated, USB reports have been sent, LEDs have been updated, and displays have been drawn.
Similar to `matrix_scan_*`, these are called as often as the MCU can handle. To keep your board responsive, it's suggested to do as little as possible during these function calls, potentially throtting their behaviour if you do indeed require implementing something special.
If the board supports it, it can be "idled", by stopping a number of functions. A good example of this is RGB lights or backlights. This can save on power consumption, or may be better behavior for your keyboard.
This is controlled by two functions: `suspend_power_down_*` and `suspend_wakeup_init_*`, which are called when the system board is idled and when it wakes up, respectively.
QMK has the ability to execute a callback after a specified period of time, rather than having to manually manage timers. To enable this functionality, set `DEFERRED_EXEC_ENABLE = yes` in rules.mk.
The first argument `trigger_time` is the intended time of execution. If other delays prevent executing at the exact trigger time, this allows for "catch-up" or even skipping intervals, depending on the required behaviour.
The second argument `cb_arg` is the same argument passed into `defer_exec()` below, and can be used to access state information from the original call context.
The return value is the number of milliseconds to use if the function should be repeated -- if the callback returns `0` then it's automatically unregistered. In the example above, a hypothetical `my_deferred_functionality()` is invoked to determine if the callback needs to be repeated -- if it does, it reschedules for a `500` millisecond delay, otherwise it informs the deferred execution background task that it's done, by returning `0`.
?> Note that the returned delay will be applied to the intended trigger time, not the time of callback invocation. This allows for generally consistent timing even in the face of occasional late execution.
The first argument is the number of milliseconds to wait until executing `my_callback` -- in the case above, `1500` milliseconds, or 1.5 seconds.
The third parameter is the `cb_arg` that gets passed to the callback at the point of execution. This value needs to be valid at the time the callback is invoked -- a local function value will be destroyed before the callback is executed and should not be used. If this is not required, `NULL` should be used.
The return value is a `deferred_token` that can consequently be used to cancel the deferred executor callback before it's invoked. If a failure occurs, the returned value will be `INVALID_DEFERRED_TOKEN`. Usually this will be as a result of supplying `0` to the delay, or a `NULL` for the callback. The other failure case is if there are too many deferred executions "in flight" -- this can be increased by changing the limit, described below.
This page used to encompass a large set of features. We have moved many sections that used to be part of this page to their own pages. Everything below this point is simply a redirect so that people following old links on the web find what they're looking for.
