quantum: Add a tap dance feature

[algernon]

With this feature one can specify keys that behave differently, based on the amount of times they have been tapped, and when interrupted, they get handled before the interrupter.

To make it clear how this is different from ACTION_FUNCTION_TAP, lets explore a certain setup! We want one key to send Space on single tap, but Enter on double-tap.

With ACTION_FUNCTION_TAP, it is quite a rain-dance to set this up, and has the problem that when the sequence is interrupted, the interrupting key will be send first. Thus, SPC a will result in a SPC being sent, if they are typed within TAPPING_TERM. With the tap dance feature, that'll come out as SPC a, correctly.

The implementation hooks into two parts of the system, to achieve this: into process_record_quantum(), and the matrix scan. We need the latter to be able to time out a tap sequence even when a key is not being pressed, so SPC alone will time out and register after TAPPING_TERM time.

But lets start with how to use it, first!

First, you will need TAP_DANCE_ENABLE=yes in your Makefile, because the feature is disabled by default. This adds a little less than 1k to the firmware size. Next, you will want to define some tap-dance keys, which is easiest to do with the TD() macro, that - similar to F(), takes a number, which will later be used as an index into the tap_dance_actions array.

This array specifies what actions shall be taken when a tap-dance key is in action. Currently, there are two possible options:

• ACTION_TAP_DANCE_DOUBLE(kc1, kc2): Sends the kc1 keycode when tapped once, kc2 otherwise.
• ACTION_TAP_DANCE_FN(fn): Calls the specified function - defined in the user keymap - with the current state of the tap-dance action.

The first option is enough for a lot of cases, that just want dual roles. For example, ACTION_TAP_DANCE(KC_SPC, KC_ENT) will result in Space being sent on single-tap, Enter otherwise.

And that's the bulk of it!

Do note, however, that this implementation does have some consequences: keys do not register until either they reach the tapping ceiling, or they time out. This means that if you hold the key, nothing happens, no repeat, no nothing. It is possible to detect held state, and register an action then too, but that's not implemented yet. Keys also unregister immediately after being registered, so you can't even hold the second tap. This is intentional, to be consistent.

And now, on to the explanation of how it works!

The main entry point is process_tap_dance(), called from process_record_quantum(), which is run for every keypress, and our handler gets to run early. This function checks whether the key pressed is a tap-dance key. If it is not, and a tap-dance was in action, we handle that first, and enqueue the newly pressed key. If it is a tap-dance key, then we check if it is the same as the already active one (if there's one active, that is). If it is not, we fire off the old one first, then register the new one. If it was the same, we increment the counter and the timer.

This means that you have TAPPING_TERM time to tap the key again, you do not have to input all the taps within that timeframe. This allows for longer tap counts, with minimal impact on responsiveness.

Our next stop is matrix_scan_tap_dance(). This handles the timeout of tap-dance keys.

For the sake of flexibility, tap-dance actions can be either a pair of keycodes, or a user function. The latter allows one to handle higher tap counts, or do extra things, like blink the LEDs, fiddle with the backlighting, and so on. This is accomplished by using an union, and some clever macros.

In the end, lets see a full example!

enum {
 CT_SE = 0,
 CT_CLN,
 CT_EGG
};

/* Have the above three on the keymap, TD(CT_SE), etc... */

void dance_cln (qk_tap_dance_state_t *state) {
  if (state->count == 1) {
    register_code (KC_RSFT);
    register_code (KC_SCLN);
    unregister_code (KC_SCLN);
    unregister_code (KC_RSFT);
  } else {
    register_code (KC_SCLN);
    unregister_code (KC_SCLN);
    reset_tap_dance (state);
  }
}

void dance_egg (qk_tap_dance_state_t *state) {
  if (state->count >= 100) {
    SEND_STRING ("Safety dance!");
    reset_tap_dance (state);
  }
}

const qk_tap_dance_action_t tap_dance_actions[] = {
  [CT_SE]  = ACTION_TAP_DANCE_DOUBLE (KC_SPC, KC_ENT)
 ,[CT_CLN] = ACTION_TAP_DANCE_FN (dance_cln)
 ,[CT_EGG] = ACTION_TAP_DANCE_FN (dance_egg)
};

This fixes #426.

[ezuk]

This is amazing! I love the implementation and how clean it is, the fact it's Quantum-style (in its own file), the use of the ranges etc. Waiting for @jackhumbert to take a look, but I think this is awesome. I think once the PR is approved I'll just take your desc above and put it right into the docs, it's great.

Thank you!

[jackhumbert]

This is awesome :) would you mind implementing it in the same way that the ones on the quantum-keypress-process branch are, and making a PR to that? Keeping all of the features like this in a folder will make things a lot easier to manage :)

On the matrix_scan_* calls, I think each of the custom matrix.c files should be calling matrix_scan_quantum - that should allow us to only have to add things to one place, and the *_kb one will still get called automatically. Not sure how I missed that before :)

[algernon]

Sure. I will update this PR tomorrow morning (CEST) latest - or in an hour, if all goes well.

Also spotted a bug in my implementation just now, so will need an update anyway: if a keymap has two tap-dance keys, pressing one after the other triggers the action for the second, instead of the first. That's a one-line fix, and I'll include it in the update too.

[algernon]

Oh wow. Github doesn't allow me to change the target branch. Closing this then, and will open a new PR soon.

[NoahAndrews]

In the future, just update the same branch. The pull request will update with the new changes when you push.

[algernon]

I know. But it won't update the target branch, as in, it will still say its to be merged into master, instead of quantum-keypress-process.

[NoahAndrews]

Oh, gotcha