1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214
/* Copyright (C) 2021 Purism SPC
* SPDX-License-Identifier: GPL-3.0+
*/
/*! The loop abstraction for driving state changes.
* It binds to the state tracker in `state::Application`,
* and actually gets driven by a driver in the `driver` module.
*
* * * *
*
* If we performed updates in a tight loop,
* the state tracker would have been all we need.
*
* ``
* loop {
* event = current_event()
* outcome = update_state(event)
* io.apply(outcome)
* }
* ``
*
* This is enough to process all events,
* and keep the window always in sync with the current state.
*
* However, we're trying to be conservative,
* and not waste time performing updates that don't change state,
* so we have to react to events that end up influencing the state.
*
* One complication from that is that animation steps
* are not a response to events coming from the owner of the loop,
* but are needed by the loop itself.
*
* This is where the rest of bugs hide:
* too few scheduled wakeups mean missed updates and wrong visible state.
* Too many wakeups can slow down the process, or make animation jittery.
* The loop iteration is kept as a pure function to stay testable.
*/
pub mod driver;
use std::cmp;
use std::time::{ Duration, Instant };
/// Carries the incoming data to affect the actor state,
/// plus an event to help schedule timed events.
pub trait Event: Clone {
fn new_timeout_reached(when: Instant) -> Self;
/// Returns the value of the reached timeout, if this event carries the timeout.
fn get_timeout_reached(&self) -> Option<Instant>;
}
/// The externally observable state of the actor.
pub trait Outcome {
type Commands;
/// Returns the instructions to emit in order to change the current visible state to the desired one.
fn get_commands_to_reach(&self, desired: &Self) -> Self::Commands;
}
/// Contains and calculates the intenal state of the actor.
pub trait ActorState: Clone {
type Event: Event;
type Outcome: Outcome;
/// Returns the new internal state after the event gets processed.
fn apply_event(self, e: Self::Event, time: Instant) -> Self;
/// Returns the observable state of the actor given this internal state.
fn get_outcome(&self, time: Instant) -> Self::Outcome;
/// Returns the next wake up to schedule if one is needed.
/// This may be called at any time, so should always return the correct value.
fn get_next_wake(&self, now: Instant) -> Option<Instant>;
}
/// This keeps the state of the tracker loop between iterations
#[derive(Clone)]
struct State<S> {
state: S,
scheduled_wakeup: Option<Instant>,
last_update: Instant,
}
impl<S> State<S> {
fn new(initial_state: S, now: Instant) -> Self {
Self {
state: initial_state,
scheduled_wakeup: None,
last_update: now,
}
}
}
/// A single iteration of the loop, updating its persistent state.
/// - updates tracker state,
/// - determines outcome,
/// - determines next scheduled animation wakeup,
/// and because this is a pure function, it's easily testable.
/// It returns the new state, and the message to send onwards.
fn handle_event<S: ActorState>(
mut loop_state: State<S>,
event: S::Event,
now: Instant,
) -> (State<S>, <S::Outcome as Outcome>::Commands) {
// Calculate changes to send to the consumer,
// based on publicly visible state.
// The internal state may change more often than the publicly visible one,
// so the resulting changes may be no-ops.
let old_state = loop_state.state.clone();
let last_update = loop_state.last_update;
loop_state.state = loop_state.state.apply_event(event.clone(), now);
loop_state.last_update = now;
let new_outcome = loop_state.state.get_outcome(now);
let commands = old_state.get_outcome(last_update)
.get_commands_to_reach(&new_outcome);
// Timeout events are special: they affect the scheduled timeout.
loop_state.scheduled_wakeup = match event.get_timeout_reached() {
Some(when) => {
if when > now {
// Special handling for scheduled events coming in early.
// Wait at least 10 ms to avoid Zeno's paradox.
// This is probably not needed though,
// if the `now` contains the desired time of the event.
// But then what about time "reversing"?
Some(cmp::max(
when,
now + Duration::from_millis(10),
))
} else {
// There's only one timeout in flight, and it's this one.
// It's about to complete, and then the tracker can be cleared.
// I'm not sure if this is strictly necessary.
None
}
},
None => loop_state.scheduled_wakeup.clone(),
};
// Reschedule timeout if the new state calls for it.
let scheduled = &loop_state.scheduled_wakeup;
let desired = loop_state.state.get_next_wake(now);
loop_state.scheduled_wakeup = match (scheduled, desired) {
(&Some(scheduled), Some(next)) => {
if scheduled > next {
// State wants a wake to happen before the one which is already scheduled.
// The previous state is removed in order to only ever keep one in flight.
// That hopefully avoids pileups,
// e.g. because the system is busy
// and the user keeps doing something that queues more events.
Some(next)
} else {
// Not changing the case when the wanted wake is *after* scheduled,
// because wakes are not expensive as long as they don't pile up,
// and I can't see a pileup potential when it doesn't retrigger itself.
// Skipping an expected event is much more dangerous.
Some(scheduled)
}
},
(None, Some(next)) => Some(next),
// No need to change the unneeded wake - see above.
// (Some(_), None) => ...
(other, _) => other.clone(),
};
(loop_state, commands)
}
#[cfg(test)]
mod test {
use super::*;
use crate::animation;
use crate::imservice::{ ContentHint, ContentPurpose };
use crate::panel;
use crate::state;
use crate::state::{ Application, InputMethod, InputMethodDetails, Presence, visibility };
use crate::state::test::application_with_fake_output;
fn imdetails_new() -> InputMethodDetails {
InputMethodDetails {
purpose: ContentPurpose::Normal,
hint: ContentHint::NONE,
}
}
// TODO: This should only test the scheduling in handle_event.
// This means it should be separated from actual application logic,
// and use a mock state instead.
#[test]
fn schedule_hide() {
let start = Instant::now(); // doesn't matter when. It would be better to have a reproducible value though
let mut now = start;
let state = Application {
im: InputMethod::Active(imdetails_new()),
physical_keyboard: Presence::Missing,
visibility_override: visibility::State::NotForced,
..application_with_fake_output(start)
};
let l = State::new(state, now);
let (l, commands) = handle_event(l, InputMethod::InactiveSince(now).into(), now);
assert_matches!(commands.panel_visibility, Some(panel::Command::Show{..}));
assert_eq!(l.scheduled_wakeup, Some(now + animation::HIDING_TIMEOUT));
now += animation::HIDING_TIMEOUT;
let (l, commands) = handle_event(l, state::Event::TimeoutReached(now), now);
assert_eq!(commands.panel_visibility, Some(panel::Command::Hide));
assert_eq!(l.scheduled_wakeup, None);
}
}