bevy_enhanced_input

[crude summit]

I also found much more compact way to write debug/technical input

type Kbd<'w> = Res<'w, ButtonInput<KeyCode>>;
type Mouse<'w> = Res<'w, ButtonInput<MouseButton>>;
// ...
        .add_systems(Update, (
            grab_cursor.run_if(|k: Mouse| k.just_pressed(MouseButton::Left)),
            release_cursor.run_if(|k: Kbd| k.just_pressed(KeyCode::Escape)),
            toggle_free_camera.run_if(|k: Kbd| k.just_pressed(KeyCode::KeyC)),
        ))

Maybe it would be worth adding to BEI some short syntax for such case?

[sharp warren]

This looks like bevy_input to me.

[crude summit]

It is, I mean that maybe we want similar short syntax in BEI, but unified for different input devices.

[sharp warren]

We have triggers for it

[thick kiln]

I updated this issue: https://github.com/simgine/bevy_enhanced_input/issues/164
I think it would still be useful for the crate to work in conditions where inputs only come from Mocking, and not from InputReader

[neat bloom]

hi. im switching our game over from leafwing to bevy enhanced input. we have a situation where we need to see if the Alt key is pressed while a Move Action is happening. i already implemented the Move action with wasd and gamepad and its working. now i need to change the behavior depending on if the Alt key is pressed. i looked through the examples and didnt see how to get more than 1 target in an observer function...

[neat bloom]

guessing i shouold use a bevy state and set that with an observer function

[sharp warren]

You can use pull-based API by basically querying the action value in a system, see docs for more details.

[sharp warren]

Makes sense. I'm open adding it, but I need @outer wyvern to merge open PRs first, otherwise this change will conflict.

[static hearth]

While updating from before the component refactor to 0.17.1, I ran into the same issue. Is there a workaround?
Before the refactor I used SwizzleAxis (SwizzleAxis also can't be attached) and Negate to bind the 3d translation of a spacecraft to a single Action like this:

actions.bind::<Translate>().to((
  Cardinal::wasd_keys().with_modifiers_each((
    SwizzleAxis::XZY,
      Negate {
        x: true,
        y: false,
        z: true,
      },
    )),
    Bidirectional {
      positive: KeyCode::Space,
      negative: KeyCode::ShiftLeft,
    }.with_modifiers_each(SwizzleAxis::ZXY),
));

These are my bindings now:

    actions!(
        ShipControl[(
            Action::<Translate>::new(),
            Bindings::spawn((
                Cardinal::wasd_keys().with((
                        SwizzleAxis::XZY,
                        Negate {
                            x: true,
                            y: false,
                            z: true,
                        }
                    )),
                Bidirectional::<Binding, Binding> {
                    positive: KeyCode::Space.into(),
                    negative: KeyCode::ShiftLeft.into()
                }
                .with(SwizzleAxis::ZXY)
            )),
        )]
    )

It crashes when SwizzleAxis and Negate aren't commented out for the Cardinal binding

[sharp warren]

The only workaround is to apply swizzle and negation on the action level.

Or don't use Cardinal

And you can also create your own preset tied to what you expect.

[static hearth]

I'll create a my own preset

[sharp warren]

You can simply copy the Cardinal code and edit it.

[neat bloom]

thanks. do you have a link to where i can read about the pull based query?

[sharp warren]

In the docs: https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/#pull-style

@thick kiln I opened https://github.com/simgine/bevy_enhanced_input/pull/184, which is based on https://github.com/simgine/bevy_enhanced_input/pull/181
You can use patch section to try it if you want.

[outer wyvern]

@sharp warren I've done a review pass on all open PRs now 🙂

A couple are blocked / waiting on author, but I've kicked off a merge for the others

[sharp warren]

Thanks!

[sharp warren]

Rebased this one

[limpid lake]

@sharp warren maybe for the ‘remove_with_requires’ issue, it can be solved with a macro like ‘remove_context!’ That will do it without any footguns?

At least for now so it will be more user friendly

[sharp warren]

I don't think it's any better then calling remove_with_requires.

[limpid lake]

Yeah fair haha

[sharp warren]

@limpid lake @thick kiln @heady mauve drafted a new release with the features you requested/submitted.

[zinc cliff]

Is there an example for how to use non-observer based actions?

migrating from 0.14 rn and struggling a bit

o, nvm just above same question

Does each action have its own entity? curious how this is actually implemented because it feels a bit like magic to me right now:
(Action::<EquipItem>::new(), EquipHotbarIndex(0), bindings![KeyCode::Digit1]),

[sharp warren]

Yes.
All macro it's basically related with BEI types. Just like children.

I highly suggest reading the quick start guide in the docs first, it's all explained.

Feel free to ask if something is unclear. I constantly try to improve docs and examples.

[zinc cliff]

Is there any event/trigger on a context becoming active/inactive or is it always necessary for a ContextActivity to be inserted to trigger the change, never modified? I see it's immutable so insert only

Trying to re-implement this:

pub fn started_flying(
    trigger: Trigger<OnInsert, ContextActivity<FlyingCamera>>,
    mut query: Query<(&Transform, &ContextActivity<FlyingCamera>, &mut FlyingState)>,
) {
    let Ok((transform, context_activity, mut state)) = query.get_mut(trigger.target()) else {
        return;
    };

    if *context_activity == ContextActivity::<FlyingCamera>::INACTIVE {
        return;
    }

    let (yaw, pitch, _roll) = transform.rotation.to_euler(EulerRot::YXZ);
    state.yaw = yaw;
    state.pitch = pitch;
}

but I'm not entirely sure if I can do this with a check either?

oh I'm dumb there is a deref to bool

[sharp warren]

Why do you check for activity?

Events won't fire for inactive contexts.

[zinc cliff]

I just want to initialize things for a specific context when it becomes active

[sharp warren]

Ah, I see

[zinc cliff]

in this case I want the flying camera to start with the same position and orientation as the previous camera

[round portal]

I thought priority means that if action is not defined in higher priority then it searches for the binding in the lower priority contexts, is that not the case?

[sharp warren]

Each context evaluated in the defined priority. Inside context all actions evaluated according to their ordering.

[round portal]

so if I have defined Action Forward with KeyW binding on context with priority 0 but not on another with priority 1 when context with priority 1 is active it means that on KeyW no Forward action will be fired?

[heady mauve]

If the priority 1 context has no KeyW binding then it should transfer through to activate the Forward action on priority 0. Unless you make the ContextActivity for the priority 0 context inactive.

[round portal]

hmm, I don't think I am removing the lower priority one, buuut...it would explain a lot. I suspect I do.

thanks!

[sharp warren]

It should fire

[heady mauve]

Are you adding the observers to entities directly, or do you have them as global observers?

[thick kiln]

How does BEI behave for entities that don't have ActionMock or Bindings; would it still trigger the Events from ActionEvents?
(this is so that when I rollback and I predict future inputs, the push-based triggers still get emitted)

Maybe what I should do is just set an ActionMock on the remote-client Action entities with their last received State/Value; that way on rollback it acts as if the State/Value stay the same, and i get updated ActionTime and ActionEvent for free

[thick kiln]

it looks like it does; it seems that BEI::Apply is independent from BEI::Update

[thick kiln]

It looks like I have to use ActionMock; because otherwise BEI::Update runs on Action entities even if they have no bindings and no mocks, and resets the State/Value to zero.

[thick kiln]

What would be the best way for me to block an Action from being updated in BEI::Update?
Adding a custom 'Filter' component to BEI? or some custom InputCondition?

[sharp warren]

Correct

[sharp warren]

Do you need this for actions from other players?

[sharp warren]

Maybe just mock the input?

[thick kiln]

i thought about that but right now it's easier if i can just exclude some actions from being updated (because of other reasons related to rollback)

besides it seems like it would be generally useful?

[sharp warren]

I think you can disable those entities with entity disabling feature

I ensured disabling works as expected, so it shouldn't break

Would that work for you?

[thick kiln]

I don't think that would work, because then my users would need to know about my custom Disabling component whenever they want to query the ActionState of remote players

[sharp warren]

Ah, you want to apply that permanently 🤔

But you previosly mentioned that you intercept actions, if you just run your logic after Update, wouldn't that work?

[thick kiln]

yes, something like that: https://github.com/simgine/bevy_enhanced_input/pull/189

i guess; but why pay the cost of updating all these entities

also then i lose the previous state of ActionState, which I rely on for predicting future states

[sharp warren]

That's fair.

But what is wrong with input mocking? Maybe we can adjust it somehow to make it compatible with rollback?
Because disabling some actions from BEI::update and doing the update manually is not the most elegant API 🤔

[thick kiln]

The issue is that I have action_predict function (that increments ActionTime/ActionEvents) that I call during rollbacks or when receiving a new InputMessage.
e.g. we had remote actions up to tick 8, current tick is 12 and we rollback to tick 4.
For ticks 4-8, we fetch the remote action from the buffer.
For ticks 8-12, we run action_predict to update ActionTime/ActionEvents.

You could say: why don't you just set the MockAction at tick 8 (during rollback) when you detect that you reached the end of the input buffer?
The reason is that I don't run BEI::Update at all during rollbacks because it's not needed and would hurt performance. I already have all the information for the inputs of all players, so I don't really want to have to call Update again

The other reason is that my API is shared between leafwing, lightyear_native, and BEI

so i cannot rely on some smart mocking API from BEI since it doesn't exist in leafwing

[sharp warren]

So you need to update specific context(s)?
Maybe just disable specific context you don't want to update?

Using https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/context/struct.ContextActivity.html

[thick kiln]

I still need the context to be active for the client-controlled entity

[sharp warren]

Can't you keep it active then?

[thick kiln]

yes the context needs to be active; but I don't want the ActionState/ActionValue to be modified for the remote-client inputs. And I don't want to run BEI::Update during rollbacks for performance reasons.

but also don't you think that this kind of filter would be useful in general?

[sharp warren]

Can't think if any possible use case other then networking 🤔

[thick kiln]

well right now you can disable an entire context, but you can't disable BEI from running for an individual Action

[sharp warren]

Alice wasn't happy when I added context disabling, but I think it's incredibly useful when you switch between contexts.
But I'm not sure how it could be useful for actions other then networking. So I want to understand the problem more to make sure the API is right.

[sharp warren]

BTW, why disable specific actions?

[thick kiln]

I disable all the Actions for remote players, not for the local player

[sharp warren]

So each player has its own context?

[thick kiln]

No the context is the same for all players. The context is called Move
But for prediction a client needs to be aware of the inputs/actions of other clients

[sharp warren]

Yeah, the type is the same, but the instance is different?

[thick kiln]

ah, yes

ah i thought context-disabling was done for the type, but it's only for the entity, i see

[sharp warren]

Yeah

[thick kiln]

the issue is that I still want apply to run, so that triggers are emitted

[sharp warren]

Ah, nvm, deactivating the context resets the values, which is not what you want.
Let me think a bit more 🤔 Disabling actions would definitely work, just not sure if it's the best way of doing this...

[thick kiln]

sure, let me know

[sharp warren]

@thick kiln what if we adjust current mocking to allow expressing "Externally mocked"?
It's basically what you did, we just need to rename ExcludeFromUpdate into something like UserMocked 😅
Another option is to somehow adjust ActionMock. Like turn it into an enum. But I think an archetype filter would work better.

[thick kiln]

yes UserMocked works for me; basically it means that i am controlling how the ActionState/Events/Time/etc. get updated for that specific Action

[sharp warren]

Yep!
Could you update the PR and include what we discussed here in the description?

[zinc cliff]

things I might suggest for docs: a mention to chords/modkeys for action bindings, and maybe of how to traverse backwards from the "target" of an action to an individual action

overall, fantastic work, ported over to 0.18 and it feels cleaner than ever 🙂

[sharp warren]

Thanks for the feedback, I'll take a look!

If you already have some ideas where and how to write about it - PRs are welcome

[thick kiln]

Removing the Context entity doesn't unregister it? I have to specifically remove ContextPriority<C> ?

[thick kiln]

yes the library is awesome

[sharp warren]

You need to call remove_with_requires::<C>

I know it's kinda unexpected, I even opened an issue on Bevy side:
https://github.com/bevyengine/bevy/issues/20753

In the meanwhile, I documented it:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/#removing-contexts

[vague temple]

@sharp warren ah meme for your troubles

[sharp warren]

@thick kiln I'll draft a new patch release once Alice approves. It's weekends, so possibly on Monday 🙂

Ah, the PR was merged because I enabled auto-merge and approved it. It's configured for one approval and CI pass 😢

[thick kiln]

thanks!

[thick kiln]

Why is despawn_related required? Doesn't linked_spawn mean that if you despawn the Context, the Actions will be despawned?

[sharp warren]

No 😢
I also found it weird, even opened this issue: https://github.com/bevyengine/bevy/issues/20252
Please, upvote if you agree 🙂

[thick kiln]

ah, but it still works if the parent entity is despawned though

[sharp warren]

Ah, if you despawn the context - actions will be despawned.

Yes, despawn_related needed only if you remove the context.

I probably should clarify this in the docs

Opened: https://github.com/simgine/bevy_enhanced_input/pull/191

[round portal]

global observers

my idea was that PlayerCtx is inserted for each player(I plan to add split screen), but before player model is spawned, you have to move through menu somehow.
Thats what ModalCtx is for. It also is for overriding some keys everytime a modal is spawned: pause, inventory, any other modal where you have to navigate, because navigating with gamepad stick right now it's not implemented and is one of the TODOs

p.s. sorry for delayed response, I'm having little time for project nowadays

[sharp warren]

Yep, the approach makes sense.

[round portal]

ok I've finally figured it out - I override wasd on the second context as well 🤦‍♂️
This has been bugging me for weeks. Oh how good it feels to carve some time for the hobby project and see the progress.

Btw, I love how the new component approach gets rid of some crutch systems I had

[sharp warren]

When you run into an issue like this next time, try to enable logging so see what's triggering and what's consuming the input.
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/#troubleshooting
This saved me a lot of time

[sharp warren]

@thick kiln the patch is up!

[vague temple]

can an action be considered started and completed in the same frame @sharp warren

Note - This is not me trying to do that

just trying to figure out where my state machien is clunka

[sharp warren]

No, here is a table:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/action/events/struct.ActionEvents.html

[vague temple]

@sharp warren would you like a meme?

[sharp warren]

Why ask, just send it 🙂

[quasi epoch]

how do i prevent errors from having spawning a bundle too large in the bind function?

@sharp warren

[sharp warren]

You can nest bundles. Like this: ((A, B), (C, D))

[spice knot]

is there a way to temporarily deactivate action contexts without deleting all the actions?

[cursive gorge]

Does entity disabling work? 🤔

[spice knot]

lemme check

[sharp warren]

Use ContextActivity<C> to disable specific context on your entity.

[sharp warren]

This also works, yes 🙂

[spice knot]

thanks

[spice knot]

when I have the same action context on 3 different entities the action value only gets updated for one of them when I press the corresponding button, is this a bug or expected behaviour?

[sharp warren]

It's expected because actions consume inputs by default.
You can configure it using ActionSettings

[native pendant]

is it possible to have bevy_enhanced_input on a headless server? for Actions replication, but without the systems that read InputReader?

thread 'main' panicked at /home/hukasu/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/bevy_ecs-0.16.1/src/error/handler.rs:141:1:
Encountered an error in system `bevy_enhanced_input::context::input_reader::update_pending`: Parameter `InputReader::keys` failed validation: Resource does not exist

[sharp warren]

Yep, it just works with the latest version.

[native pendant]

0.18?

[sharp warren]

Yes

I linked it in the opened issue.

[native pendant]

nice

[coarse brook]

I have two contexts, let’s call them Base and ObjectGrabbed, where ObjectGrabbed has a higher priority. Both of them have an action on left mouse click: Select for Base and PlaceObject for ObjectGrabbed. When the object is placed, the ObjectGrabbed context is cleaned up.

When left clicking and both contexts are active, PlaceObject is obviously called. For some reason the Base context still gets Select called when ObjectGrabbed is despawned, even though Select is configured to require reset.

What’s the best way to block this? I want Select to only be triggered once the user has let go of the mouse button after placing the object.

[sharp warren]

Check this example: https://github.com/simgine/bevy_enhanced_input/blob/master/examples/context_layering.rs

[heady mauve]

Does PlaceObject also have require reset?

[coarse brook]

Aaaah thank you! I had already set up require reset in other things where it worked, but I had it conceptually the other way around. Now that I added require reset to PlaceObject as well it works as expected. Phew, I thought I'd have to add a hacky workaround for now. 😄

[vague temple]

@sharp warren Should an input buffer be unreliable?

[sharp warren]

You mean for networking?

[vague temple]

yes

[sharp warren]

If you use Lightyear, it already handles the input for you. It should be handled in a bit special way.
But yeah, under the hood it's unrealiable.

[vague temple]

that mean packet losses can affect it?

[sharp warren]

Yes, but it's handled with the mentioned special way 🙂
Check this video starting from this timestep if you need details: https://youtu.be/W3aieHjyNvw?feature=shared&t=1528

[vague temple]

okay

[sharp warren]

I believe that's what lightyear_inputs and rewind_input do.

[vague temple]

interesting this although avoids some issue can still scre my char controller

[sharp warren]

If you use Lightyear's BEI integration and its prediction/rollback API, you shouldn't have any issues with your character 🤔
It's either a bug in the crate or in how you use it.

[cursive gorge]

Yep, I just send things unreliable, but send a whole array of ticks, all of the ones that might still arrive in time on the server, and make sure the client is far enough ahead to handle at least 1 lost packet (though if inputs are very important this could be increased, or you might just want to go for reliable events in the cases where they need to definitely happen, even if not on the desired tick)

[vague temple]

That is my case

If a single tick is lost there is a slight chance a state may never be removed

[alpine socket]

... or you might just want to go for reliable events in the cases where they need to definitely happen, even if not on the desired tick)

This is how we handle it, reliable unordered network triggers for infrequent, but important state changes, and unreliable player movement.

[quasi epoch]

i'm making a 0.17 migration for bevy_enhanced_input, i've fixed almost everything but the one thing remaining is it tries to spawn things with various direction types or something? i'm not really sure what's going on, i tried searching the migration guides but couldn't find anything on it

error[E0185]: method `spawn` has a `self` declaration in the impl, but not in the trait
  --> src/preset/ordinal.rs:87:5
   |
87 |     fn spawn(self, world: &mut World, entity: Entity) {
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ `self` used in impl
   |
   = note: `spawn` from trait: `fn(MovingPtr<'_, Self>, &mut bevy::prelude::World, bevy::prelude::Entity)`

error[E0599]: no method named `spawn` found for struct `cardinal::Cardinal<N, E, S, W>` in the current scope
   --> src/preset/ordinal.rs:94:18
    |
 94 |         cardinal.spawn(world, entity);
    |                  ^^^^^ this is an associated function, not a method
    |
   ::: src/preset/cardinal.rs:11:1
    |
 11 | pub struct Cardinal<N, E, S, W> {
    | ------------------------------- method `spawn` not found for this struct
    |

[quasi epoch]

@inner sky

[sharp warren]

Looks like the signature changed a little:
https://dev-docs.bevy.org/bevy/ecs/spawn/trait.SpawnableList.html#tymethod.spawn

Try writing this as this: MovingPtr<'_, Self>

[quasi epoch]

SpawnableList::spawn(move_as_ptr!(xy), world, entity);
this works

well, at least it doesn't throw an error

nvm

that macro doesn't exist

ugh these moving pointer things are confusing

[sharp warren]

Ah, looks like we also need https://github.com/bevyengine/bevy/pull/20984 to land.

On second thought, no 🤔
The problem is that when we create a spawnable list, we need to convert it into MovingPtr.

[sharp warren]

Looks like it exists: https://dev-docs.bevy.org/bevy/ecs/ptr/macro.move_as_ptr.html

[sharp warren]

I think you just need to call move_as_ptr!(xy), on a separate line. This will turn xy into a MovingPtr. And then you should be able to xy.spawn(..)

See the example here: https://dev-docs.bevy.org/bevy/ecs/ptr/struct.MovingPtr.html#method.move_field

[inner sky]

Redirected to #1404237363550486751

You should definitely not have to puzzle this out on your own 🙂

[shrewd wolf]

    fn spawn(this: MovingPtr<'_, Self>, world: &mut World, entity: Entity) {
        let (n, ne, e, se, s, sw, w, nw) = unsafe {
            deconstruct_moving_ptr!(
                this => {
                    north,
                    north_east,
                    east,
                    south_east,
                    south,
                    south_west,
                    west,
                    north_west,
                }
            );
            (
                north, north_east, east, south_east, south, south_west, west, north_west,
            )
        };
        let cardinal = Cardinal {
            north: n.read(),
            east: e.read(),
            south: s.read(),
            west: w.read(),
        };
        move_as_ptr!(cardinal);
        SpawnableList::spawn(cardinal, world, entity);

        world.spawn((BindingOf(entity), ne.read(), SwizzleAxis::XXZ));
        world.spawn((BindingOf(entity), se.read(), SwizzleAxis::XXZ, Negate::y()));
        world.spawn((
            BindingOf(entity),
            sw.read(),
            SwizzleAxis::XXZ,
            Negate::all(),
        ));
        world.spawn((BindingOf(entity), nw.read(), SwizzleAxis::XXZ, Negate::x()));
    }
``` this might work? This is somewhat unergonomic though lol

[sharp warren]

Looks awful 😅
I glad it's only internals, but still...

[sharp warren]

But why deconstruct_moving_ptr is needed?

I would expect to see only move_as_ptr!

[shrewd wolf]

because it's moving bundles within bundles

[sharp warren]

Ah, right, this is MovingPtr

[shrewd wolf]

there's definitely room for improvement here in terms of usability though..

[sharp warren]

Could you report it to the https://discord.com/channels/691052431525675048/1404237363550486751 ?
Since we introduced it in this release, maybe we could think of some possible improvements? 🤔

[shrewd wolf]

yeah but it's a tough problem I think

[analog fiber]

Hi, I’m trying to set up an input action that only triggers when the middle mouse button is held down and the mouse is dragged (so MMB + motion)

I didn’t see an example of this in the docs — what would be the recommended way to implement it?

[heady mauve]

I would probably create an action when the middle mouse button is held and create a Chord condition that uses the held button action for the dragging. Create the held action first, then the action with the Chord condition and finally any other actions that use mouse motion afterwards to make sure they are prioritized correctly. Although if you use any mouse motions with keyboard modifiers you might have to make a separate context that you switch to instead.

[analog fiber]

Got it, thanks! But it does feel a lot more involved than what I’m used to with leafwing_input_manager though

[analog fiber]

it doesn’t work for me, it just panics. Maybe I wired it wrong?

[heady mauve]

Yeah, you are combining trying to create a Context and Actions in a Bundle with using command spawning to try and create the Context and different Actions.

[heady mauve]

You might want to look at the all_conditions example to see how you should create all the actions of your MainCamera context via spawn commands.

[analog fiber]

it says: Bundle ... has duplicate components: bevy_enhanced_input::action::relationship::Actions<MainCamera>

So like, how do you mix actions! and SpawnWith together?

[heady mauve]

Did you look at the all_conditions example? It doesn't use the actions! macro.

[analog fiber]

Ahh, so this needs a completely different declaration syntax then

[heady mauve]

Yeah, I'm not sure if Chords can be done in a Bundle since they have to hook into other Actions.

[analog fiber]

Weird, the result of MMB drag is: 2025-09-14T06:38:02.036267Z DEBUG Fired CameraOrbit with value [0, 0]

MMB drag gets triggered with zero values, but the R/F/Q/E keys work fine

🤔 Seems similar to an old leafwing_input_manager issue where chords with axial values were ignored

Or maybe I missed some docs, or is it actually intended to work this way?

[sharp warren]

You created 3 actions.

1. Mouse move
2. Mouse middle
3. Action that triggers when both are active.
   It's expected for it to be zero since you didn't assign any binding that would drive the value.

If you want to get the value, you need to create only 2 actions and attach the chord condition to the action value of which you want to read.

For example:

1. Mouse middle aciton
2. Mouse move action, with a chord to the mouse middle action. You will get the mouse move action.

[analog fiber]

Thanks! It works, but it's giving a [0, 0] when I hold down the middle mouse button without moving the mouse 😂

[vague temple]

/// Input context for player actions. This is here due to the fact player inputs are sometime used in other contexts
///
/// ## Actions
/// [`MoveInput`], [`RotateInput`], [`JumpInput`], [`DashInput`], [`Attack`],
/// [`NextWeapon`], [`PreviousWeapon`].
#[derive(Component, Serialize, Deserialize, Reflect, Clone, Debug, PartialEq)]
pub struct PlayerInputs;

impl PlayerInputs {
    /// NOTE DO NOT TOUCH THIS ORDERING OR YOU WILL FUCK UP DETERMINISM
    pub fn context_player() -> impl Bundle {
        (
            Self,
            actions!(Self[
                // MOVE INPUTS
                (
                    Action::<MoveInput>::new(),
                    DeadZone::default(),
                    Bindings::spawn((
                        Cardinal::wasd_keys(),
                        Axial::left_stick(),
                    )),
                ),
                (Action::<RotateInput>::new(),ActionMock::new(ActionState::Fired, Vec2::ZERO, MockSpan::Manual)),
                (Action::<JumpInput>::new(),bindings![KeyCode::Space]),
                (Action::<DashInput>::new(),bindings![KeyCode::ShiftLeft]),
                // GUN INPUTS
                (Action::<Attack>::new(),bindings![MouseButton::Left]),
                (Action::<NextWeapon>::new(),bindings![KeyCode::KeyE]),
                (Action::<PreviousWeapon>::new(),bindings![KeyCode::KeyQ]),
                (Action::<Reload>::new(),bindings![KeyCode::KeyR]),
                (Action::<Interact>::new(),bindings![KeyCode::KeyT]),
                // BET INPUTS
                (Action::<IncreaseAmount>::new(),bindings![KeyCode::KeyH]),
                (Action::<DecreaseAmount>::new(),bindings![KeyCode::KeyN]),
                (Action::<IncreaseThreshold>::new(),bindings![KeyCode::KeyK]),
                (Action::<DecreaseThreshold>::new(),bindings![KeyCode::KeyL]),
            ]),
        )
    }
}```

Question how can i split the bundle (make sub bundles so i dont get too many components error) in this macro? I know this sounds weird but the moment I add a comma I get a weird duplicate components error

Is it by writing another actions?

[analog fiber]

That code already adds PlayerInputs and Actions::<PlayerInputs>

When you also add another bundle with actions!(Self[..]), it inserts another Actions::<PlayerInputs>, so you get a duplicate

[inner sky]

pretty sure I saw someone else have the exact same problem a while ago

But I don't remember the solution

[sharp warren]

I think you can attach Down condition 🤔

Additionally

[analog fiber]

Ahh I just remembered I've seen this before. Sorry for the trouble

[sharp warren]

Hm... How it's usually solved when you need to spawn many children?

It's the same limitation.

You could do SpawnWith instead of using the actions! macro, but it will be ugly.

[vague temple]

I am confused what do you mena

[inner sky]

spawning >16 children with children! doesn't work, so you need to use with_children (or similar)

[vague temple]

i didnt know that was a thing

[sharp warren]

Same here, but with_related::<Actions<Self>>

I think it's solved on main

And it will just work after the update to Bevy 0.17.

[vague temple]

So in thesis my code would just work after 0.17 and next bei update? Well that works for me

[vague temple]

Yeah I just saw the api in the example it quite stinks

[sharp warren]

Correct. No changes in BEI will even be needed to support it, it just start to work.

[vague temple]

marvelous

[analog fiber]

🤔 Got a weird issue: I have MMB drag and RMB drag actions, but if I add Down to both, one action just never fires (for me CameraOrbit never works, doesn’t matter which mouse button or order of spawning), even with Down(0). If I remove Down, they both work (though with 0 values).

let rmb_held = context
    .spawn((Action::<RmbHeld>::new(), bindings![MouseButton::Right]))
    .id();
context.spawn((
    Action::<CameraRotate>::new(),
    Bindings::spawn((
        Spawn((
            Binding::mouse_motion(),
            Down::new(0.0),
            Chord::single(rmb_held),
        )),
    )),
));

let mmb_held = context
    .spawn((Action::<MmbHeld>::new(), bindings![MouseButton::Middle]))
    .id();
context.spawn((
    Action::<CameraOrbit>::new(),
    Bindings::spawn((
        Spawn((
            Binding::mouse_motion(),
            Down::new(0.0),
            Chord::single(mmb_held),
        )),
    )),
));

[sharp warren]

Actions consume inputs by default.
Insert ActionSettings where mouse motion won't be consumed.

On second though, consumption happens only when the action activates 🤔

Try enabling trace-level logging to see what's actually happening:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/#troubleshooting

[analog fiber]

Thanks a lot, that fixed it!

Really nice lib, but yeah it’s pretty different from leafwing_input_manager

so in the migration I’ve been going back and forth between the docs and the few examples

Any plans to add more examples?

[analog fiber]

Yeah, CameraRotate consumed the input so CameraOrbit couldn't get the value

[sharp warren]

I want to include this one: https://github.com/simgine/bevy_enhanced_input/pull/173
If you have more ideas - feel free to PR. But I prefer to have runnable examples that showcase different patters. And doc examples for small things.

[ionic sundial]

What's the best way to deal with actions! macro having a limit of 12 actions in it? I have more than 12 actions to add. I guess, just don't use the macro?

[inner sky]

Yeah use with_related::<Actions<Self>>

[inner sky]

You sure? 0.17 moves the bundles off the stack while spawning them, but the type system still needs a hardcoded limit without variadics AFAIK

[sharp warren]

I remember there was a PR specifically for related and children macro to autonest the values

[inner sky]

that would be hella nice

[sharp warren]

And since we use related! inside actions! it should just work after 0.17 🙂

@quasi epoch how the migration to 0.17 is going? Maybe you could open a draft with what you have?

[inner sky]

not possible

move_as_ptr! is broken upstream

sec

[sharp warren]

Oh, I didn't know 😢

[inner sky]

https://github.com/bevyengine/bevy/issues/21050

[sharp warren]

Found!
https://github.com/bevyengine/bevy/pull/18865

[inner sky]

I think you could work around it by importing bevy_ptr

since I believe that's what the macro points to

[inner sky]

Oh wow! Is that in the release notes?

It is not!

opening an issue

this is too awesome to not mention

[quasi epoch]

i tried doing that, it didn't seem to work

maybe i did something wrong tho ill try again tomorrow

[vague temple]

But is less pretty 😫

[inner sky]

Wait for 0.17 then :p

[vague temple]

How dare ye use logic against my pretty abstract guideline

[sharp warren]

You don't have to finish it. Even if you open a PR with what you have, it's already helpful for me - I'll finish the migration 🙂

[quasi epoch]

alright, ill do it in about 5 hours when i get home

[quasi epoch]

@sharp warren opened

[sharp warren]

Thank you!

[quasi epoch]

@inner sky i've fixed almost everything, but i've just got this left

error[E0185]: method `spawn` has a `self` declaration in the impl, but not in the trait
  --> src/preset/axial.rs:44:5
   |
44 |     fn spawn(self, world: &mut World, entity: Entity) {
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ `self` used in impl
   |
   = note: `spawn` from trait: `fn(MovingPtr<'_, Self>, &mut bevy::prelude::World, bevy::prelude::Entity)`

about 5 of those

impl<X: Bundle, Y: Bundle> SpawnableList<BindingOf> for Axial<X, Y> {
    fn spawn(self, world: &mut World, entity: Entity) {
        let xy = Axial {
            x: self.x,
            y: self.y,
        }
        .with(SwizzleAxis::YXZ);
        move_as_ptr!(xy);
        SpawnableList::spawn(xy, world, entity);
    }

    fn size_hint(&self) -> usize {
        2
    }
}```

[inner sky]

can you replace self with MovingPtr<'_, Self>,?

[quasi epoch]

just more errors

do you wanna see them?

[inner sky]

I don't think I can help much there, but @swift grail can 🙂

[swift grail]

Please do share.

[sharp warren]

You need to unpack the values.

See this message: #1297361733886677036 message

[ionic sundial]

It's cool how BEI supports Disabled OOTB. No idea if the below is "best practice" but I realised I can just disable my actions when the game is paused.

The other option (which I was originally doing) was checking the state at the top of my observer function. Observers don't support run_if like systems (which I still think could be neat), so I made all of my action observer functions check they were in the right state (mostly just to make sure the game wasn't paused).

Once I realised Disabled works, I added support for DisabledOnEnter and EnabledOnExit to my game and I think I like this better.

// ...
commands.entity(book_player_entity).insert((
    TroopRosterPlayer,
    ContextPriority::<TroopRosterPlayer>::new(**book_player_context_priority + 1),
    actions!(TroopRosterPlayer[
        (
            Action::<PreviousRegimentAction>::new(),
            bindings![KeyCode::KeyW],
            DisabledOnEnter(CampaignPaused::Active),
            EnabledOnExit(CampaignPaused::Active),
        ),
        (
            Action::<NextRegimentAction>::new(),
            bindings![KeyCode::KeyS],
            DisabledOnEnter(CampaignPaused::Active),
            EnabledOnExit(CampaignPaused::Active),
        ),
    ]),
));

[outer wyvern]

This was my hope 😄

[sharp warren]

Yes, this is intentional, and we have tests for it.

[thick kiln]

So using Disabled is equivalent to disabling the Context?

[sharp warren]

Yeah, disabling will disable all contexts on the entity.

[ionic sundial]

@sharp warren @outer wyvern re https://github.com/simgine/bevy_enhanced_input/issues/194 (Depend on subcrates directly not bevy in actions! and bindings! macros), I discovered through a post on #engine-dev that it looks like an accepted solution in the Rust ecosystem here is to use a __private mod.

Both serde 1 2 and bitflags 3 do this.

I patched BEI locally and tested it:

• With my game which uses bevy_* subcrates and I was successfully able to remove the unneeded bevy dep from my Cargo.tomls and my game was able to compile.
• With the examples in BEI which use bevy, i.e., there's no regression.

Would you accept a PR?

The diff is tiny:

diff --git a/src/action/relationship.rs b/src/action/relationship.rs
index 97cd59b..6009dd9 100644
--- a/src/action/relationship.rs
+++ b/src/action/relationship.rs
@@ -166,6 +166,6 @@ pub type ActionSpawnerCommands<'w, C> = RelatedSpawnerCommands<'w, ActionOf<C>>;
 #[macro_export]
 macro_rules! actions {
     ($context:ty [$($action:expr),*$(,)?]) => {
-        ::bevy::prelude::related!($crate::prelude::Actions<$context>[$($action),*])
+        $crate::__private::related!($crate::prelude::Actions<$context>[$($action),*])
     };
 }
diff --git a/src/binding/relationship.rs b/src/binding/relationship.rs
index 696fb84..b8a5e24 100644
--- a/src/binding/relationship.rs
+++ b/src/binding/relationship.rs
@@ -96,7 +96,7 @@ pub type BindingSpawnerCommands<'w> = RelatedSpawnerCommands<'w, BindingOf>;
 #[macro_export]
 macro_rules! bindings {
     [$($binding:expr),*$(,)?] => {
-        ::bevy::prelude::related!($crate::prelude::Bindings[$($crate::prelude::IntoBindingBundle::into_binding_bundle($binding)),*])
+        $crate::__private::related!($crate::prelude::Bindings[$($crate::prelude::IntoBindingBundle::into_binding_bundle($binding)),*])
     };
 }
 
diff --git a/src/lib.rs b/src/lib.rs
index 41be7c7..f1fac64 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -435,6 +435,13 @@ pub mod prelude {
 
 use bevy::{input::InputSystem, prelude::*};
 
+// Private module for macro internals.
+#[doc(hidden)]
+pub mod __private {
+    pub use bevy::prelude::related;
+}
+
 use condition::fns::ConditionRegistry;
 use context::{
     ContextRegistry,

[outer wyvern]

I would accept this with extremely clear doc comments about what's going on and why

[brittle swallow]

I just re-export whatever items I need where the macro is defined (https://github.com/CorvusPrudens/bevy_seedling/blob/master/src/pool/sample_effects.rs#L123).

I never liked private modules. But that’s certainly a valid way to do it!

[sharp warren]

This is ugly 😅 But from what I understand, the crate will depend on bevy this way, right?

I mean, yeah, your game will be able to use bevy_*. But what the point if the crates you use depend on bevy?

[brittle swallow]

Hm, shouldn’t BEI move to depending on subcrates anyway in anticipation of upstreaming?

[sharp warren]

Yes. Once it upstreamed, it will just work with both bevy and bevy_*. But until it's not upstreamed, it can't nicely work with both.

[brittle swallow]

It can’t?

Regardless, hoping bevy is a meaningful crate path for downstream users is less robust than reexporting the items you need. Whether you care about that case, like for users who use subcrates (even if bei doesn’t), is of course up to your discretion.

[sharp warren]

Yep, I pointed above that the mentioned workaround will result in using bevy anyway.

[brittle swallow]

Yes, but it won’t be in scope in the user’s crate. So it won’t compile.

[sharp warren]

If BEI depends on bevy, there are no advantages in using bevy_* for users.

We could also just depend on bevy_*, but then it won't work for users who use bevy

[brittle swallow]

People who depend on subcrates may do it for a number reasons. For one, it makes it easier to keep the number of dependencies down. I assume BEI doesn't depend on bevy with default features, yes? If so, I don't see the point in preventing these subcrate users from using BEI. Wether some crate depends on bevy directly really doesn't matter.

[sharp warren]

I think Bevy should document this. I see people ask this a lot. There are disadvantages with both approaches.

1. bevy - longer compile times because all third-party crates need to wait for all Bevy sub-dependencies to compile. And it pulls more, such as bevy_ecs, and bevy_math, while you might need only the ECS part.
2. bevy_* - no mentioned problem with compile times. But users need to manually enable all necessary deps on bevy. So if a third-party plugin requires some stuff, you also need to manually enable it on bevy, otherwise it won't appear in prelude and DefaultPlugins, while being in your dependency tree. It also completely breaks patching, you need to patch every single bevy subcrate in your tree because of this.

I'd recommend a balanced approach:

• Crates that can be used outside of Bevy and depend only on certain components should depend on bevy_*.
• Crates that make sense only with Bevy should depend on bevy. I'd trade fresh compile times for patching, prelude and DefaultPlugins with depds that enabled automatically any time. Especially for games, where you pull all this stuff anyway.

[brittle swallow]

Well, really we should fix the... rather unpleasant plugin situation (crates should be able to register the plugins they depend on in an idempotent way). But yes, 2 can problematic now.

[sharp warren]

And develop tools for patching. It's very common to patch your game when you need a fix or some feature and don't want to wait for the next release.
Doing this manually is ugly: https://github.com/simgine/simgine/blob/4e9947b33cf95f915d9fc82eb6ac42e43a72aaee/Cargo.toml#L47

[ionic sundial]

Yeah in this case I’m not thinking of reducing dependency compile times by talking about BEI not depending on bevy. That could just be a separate discussion to this maybe.

The discussion specifically this time for me is just “how can I make my game compile with OOTB BEI without forcing my game to depend on bevy”.

I know I can just stick a bevy dep in my crate to make it compile (and I’ll just continue to do that if I have to), but it would be nice if BEI (and bevy_asset_loader) worked out of the box for both these cases (and I believe the patch above does this—if there are other ways to get this to work as Corvus suggests I’m down with that too).

[sharp warren]

I think it's better to add bevy dependency to the game, rather then adding a workaround to BEI to make it compile with bevy_* 🤔 You won't get any benefits from it this way.

[brittle swallow]

Not having to do an extensive refactor is a pretty good benefit! That's part of what I was trying to get at.

(Moving to and from subcrates / bevy is very annoying in any nontrivial project.)

[ionic sundial]

There’s also other non quantifiable benefits. Eg I have my game broken down into crates and I can clearly see the dependency tree those depend on. If I have a giga ‘bevy’ dep in a crate does it depend on render? ECS? State? UI?

So subcrates isn’t just about compile times it’s about code architecture and cognition.

[sharp warren]

Yeah, it's definitely annoying, but I'd highly recommend doing so. Because in the end you will end up with a setup that is easier to maintain.

[sharp warren]

Yes, cargo tree will output only bevy. But it's all here in Cargo.toml. And you need to only enable the features your crate needs. Third-party crates will enable necessary things automatically.

[ionic sundial]

Anyway my goal wasn’t to raise this again, it was just to suggest a possible fix for my original issue. But if it’s not desired I won’t create a PR with that patch.

[sharp warren]

Yeah, just wanted to express my thoughts on it for others 🙂

[slim field]

just want to say that bevy_enhanced_input is really nice and imo a clear improvement over leafwing input manager so far

[sharp warren]

Thanks, I'm glad you like it 🙂

[inner sky]

@sharp warren @quasi epoch I'm blocked by the BEI migration. If you both don’t have anything against it, I'll try to finish the existing PR today

[sharp warren]

It would be awesome!

Just quite busy with work and Replicon right now 🙂

[quasi epoch]

yeah, go ahead, i don't really understand the moving ptr stuff 👍

[inner sky]

@sharp warren may I rename EnhancedInputSet to the new official convention, i.e. EnhancedInputSystems?

[sharp warren]

Sure!

[inner sky]

@sharp warren bevy::prelude::Press now clashes with bevy::prelude::Press

same for Release

Context: the Bevy types used to be called Pressed and Released

[sharp warren]

Not again 😢

[inner sky]

And another thingy

#[derive(EntityEvent)]
pub struct Started<A: InputAction> {
    /// Entity that this event was triggered on.
    #[event_target]
    pub context: Entity,

    /// Action that triggered this event.
    pub action: Entity,

    /// Current action value.
    pub value: A::Output,

    /// Current action state.
    pub state: ActionState,
}

Started and friends now want to have their event target as a field

I called it context for now

but I could also just call it entity

[sharp warren]

I remember renaming it to avoid collision with picking, but it's renamed again 😢
Probably we need to rename them into JustPressed and JustReleased...

[sharp warren]

How about target?

[inner sky]

eeeeh I don't like that

we just migrated away from .target

on Bevy itself

So this feels very 0.16 to me

Bevy itself advocates .entity where possible

I only picked .context because there are two entities here

[sharp warren]

No way! I their code they almost never use entity for Entity

[sharp warren]

Let's go with context 🙂

[inner sky]

yes way: https://github.com/bevyengine/bevy/blob/v0.17.0-rc.2/release-content/migration-guides/observer_and_event_changes.md

But yeah I know, it's usually .id() in Bevy APIs 😄

[sharp warren]

It's also sometimes just the name of the thing.

[inner sky]

yeah you're right. let me rephrase: Bevy advocates <some_action>.entity for what used to be trigger.target()

[sharp warren]

Ah, got it.

I'd advocate for this.
I.e.:

#[derive(EntityEvent)]
struct Explode {
    player: Entity
}

[inner sky]

#[derive(EntityEvent)]
struct Explode {
    #[event_target]
    player: Entity
}

yeah makes sense 🙂

[sharp warren]

I.e. when it makes sense. When it could be anything, I usually go with just entity, yeah

[inner sky]

Also, you're nudged to no longer use generic trigger names

e.g. instead of trigger: Trigger<OnAdd, Foo> it's add: On<Add, Foo>

I'm also doing that in our examples

[sharp warren]

Agree, I like it.

Reads better

[inner sky]

I have one issue with .context

Looking at the examples, it comes up a lot

and the Bevy API just uses .entity

so this may be a conflict for beginners

since they're used to .entity

[sharp warren]

Since we have 2 entities inside, I'm more inclined to context.

Having action and entity is a bit confusing.

[sharp warren]

I think this makes sense when entity is something generic. Like for a pointer API.

[inner sky]

Alright, makes sense to me 🙂

And JustPressed for Press

it's supposed to be present tense

[sharp warren]

Yeah, looks like we have to rename 😢

[inner sky]

but I don't have a better suggestion

PressNow

meh

PressForRealForReal

[sharp warren]

😅

[inner sky]

StartPress I guess

no wait

that doesn't make sense

JustPress sounds awful

[sharp warren]

Tough choice

[inner sky]

Oh I probably should also rename Fired to Fire

@sharp warren is that fine?

DW I'm adding deprecated aliases everywhere

[sharp warren]

In Unreal it was actually Triggered. I went with Fired to avoid conflicts with Bevy Trigger 😅

[inner sky]

Also, for move: On<Fired<Move>> I need to call it move_ because move is a keyword and r#move is extremely ugly

[inner sky]

Oh hey you wanna go with Trigger now?

wait no, that clashes with the deprecated type export in Bevy

But 0.18 could have Trigger 😄

[sharp warren]

I think Fire is nicer. It's shorter 🙂

[sharp warren]

Maybe keep the name and remove from prelude? 🤔

[inner sky]

That would work, if the ergonomics hit is tolerable

@outer wyvern opinion?

[sharp warren]

Maybe rename the Move action from examples into something else? 🤔

[inner sky]

I think it's better to leave it, as I expect many real-world projects to have named their action Move

(me included)

connection_messages: MessageReader<GamepadConnectionEvent>

Baaah I hate the Messages that are still called Event

This is not a bug, Cart left it intentionally

But it reads awfully

I'll rename that part of the example to just use connections and connection instead of the confusing _event suffix

[sharp warren]

Yeah, I usually don't use suffixes like this

[inner sky]

In the vein of Fired -> Fire, can I also do Started -> Start?

And Canceled -> Cancel

Ooh and also Completed -> Complete

I sure hope those don't conflict too much with user types

[sharp warren]

Yep. Looks like Bevy now uses this naming convention.

[inner sky]

exactly

[sharp warren]

I feel like users rarely name their structs in present-tense, so I think it's fine 🙂

Unless they named their buttons like this 🤔

[inner sky]

bevy_picking strikes again

They also use Cancel

[sharp warren]

Oh no

[inner sky]

Oh no indeed

[sharp warren]

Let's continue with present tense. Yeah, there is a bit of a naming collision, but keeping Cancelled won't be that much better.

Maybe once upstreamed we implement picking events as BEI actions.

[inner sky]

same for Press and Release?

[sharp warren]

Yep.

[sharp warren]

I.e. having 2 types Cancelled from BEI and Cancel from picking aren't less confusing than just having a name collision. Better to just ask users to write bei::Cancel and picking::Cancel if they happen to be used in the same module.

[inner sky]

Do you want me to keep those in our prelude?

[inner sky]

fully agreed

Also, it's just pushing the problem down the road given that we want to upstream BEI

[sharp warren]

If we keep them in prelude, how it will work for users who imported both preludes with *? 🤔

[inner sky]

use bevy::prelude::*;
use bevy_enhanced_input::prelude::{*, Press, Release, Cancel}

[sharp warren]

Yeah, makes sense, but I mean this will still compile if no Press, Release, etc are used?

use bevy::prelude::*;
use bevy_enhanced_input::prelude::*;

[inner sky]

yep 🙂

Rust is smart enough about that

[sharp warren]

What do you think about adding a small note about it in the docs? Just mention bevy_picking and the whole upstreaming situation.

[inner sky]

I added it to the changelog already

I could also add it to the docs of the individual items though

sounds like a good idea 🙂

[sharp warren]

Let's add it to the docs, yeah.

[inner sky]

mind running the CI?

I also bumped the version of the macro crate even though there are no changes FYI

so that they stay in version lockstep

[sharp warren]

Done

[inner sky]

hehe, hotbar is not looking so hot

oof something broke no_std

And a test fails locally

@sharp warren any idea what may have caused this?

[sharp warren]

I think it's non-UI example, it's fine 🙂

[sharp warren]

Weird. I wouldn't expect any of the condition tests to fail.

Maybe something changed related to time in Bevy?

Tap is timing-related

[inner sky]

I don't recall reading anything about that

[inner sky]

ah, nice

I'll chase the no_std thing for now

[sharp warren]

I'd check timings in the tests 🤔

[inner sky]

mind running the ci again?

I think I'll open an issue upstream about the no_std thing

don't think that's something the migration introduces

[sharp warren]

Strange that I have to click it every time until a single Pr is merged from you :(

[inner sky]

It's running the CI from my local changes, so in theory, I could have added a bitcoin miner to the CI in my PR 😛

[sharp warren]

Can't you do it anyway after one merged PR to the repo? 🤔

[inner sky]

yeah but IIRC this was a real scenario that GitHub had to face

People created throwaways and then just submitted PRs automatically everywhere

[sharp warren]

Yeah, I remember 😅
I guess you can't farm much if you open PR's only to the repositories you contributed

[inner sky]

@sharp warren is it intentional that we always depend on serialize?

That seems wrong

[sharp warren]

No, just never bothered to remove the dependency 😅

Yeah, worth fixing.

[inner sky]

I think that's it actually

I think the serialize feature unintentionally always activates bevy_render

[sharp warren]

Ah, a missing ?

See, my laziness helped us 😅

[sharp warren]

I think the same applies to transform 🤔

And time

[inner sky]

those are not optional

[sharp warren]

Ah, okay

[inner sky]

For now, I'm fixing the issue by just not testing the new serialize feature in no_std

Mind running the CI again?

If I did it right, only the failing test should remain

which honestly puzzles me a bit

[sharp warren]

Yeah, and it's only a single test? Weird...

[inner sky]

mind assigning GitHub Copilot for a review? It's almost certainly not gonna help, but IME it's good at spotting trivial stuff like "you accidentally deleted a !"

nvm found it

[sharp warren]

What was the issue? 😅

[inner sky]

spot the error 😄

Alright, migration done 👍

@sharp warren I'm doing a quick break, but I can work through any issues you have with the migration when I'm back

In case you don't have any issues, I'll be migrating Foxtrot later

[sharp warren]

Ah, the timer check was replaced with a variable 😅

[sharp warren]

Thanks, I'll take a look soon!

[inner sky]

bingo 😄

[sharp warren]

Looking into the PR right now

@inner sky what do you think about Move -> Movement?

[outer wyvern]

Yes please

[inner sky]

Eeh, I personally would stay with the brief version, but I can change it 🙂

[sharp warren]

I'll do it myself 🙂

I also like the shortness 🤔

[sharp warren]

I'm just in the process of the review

[sharp warren]

Will change to Movement.
Just to avoid colliding with the move keyword 🙂

[inner sky]

I need to fix the CI anyways

so I'll do it real quick

[sharp warren]

Okay, I'll rebase :)

I'll make a few tiny changes, such as Move -> Movement.

You did a really good job, this saved me a ton of time. I also like your taste in naming 🙂

[inner sky]

thanks!

[inner sky]

just pushed the rename, sorry 😅

[sharp warren]

No problem, I'll just drop my commit

[sharp warren]

You missed one in the readme 😛

[inner sky]

hehehe

done

[sharp warren]

Pushed the changes. I think I'm ready to draft a new release if there are no suggestions about the changes.

I'm thinking about keeping it in a branch untill the release drops, so the latest master points to the latest Bevy release.

Otherwise users sometimes get confused.

[outer wyvern]

I'll review quick, but I expect this to be fine

[sharp warren]

I'll wait for your review 🙂

[outer wyvern]

LGTM: kicked off a merge

[sharp warren]

Managed to quickly retarget to bevy-0.17 🙂

Published

[inner sky]

I just migrated Foxtrot and it looks like the migration is working as expected: BEI works, and the migration itself uses the deprecated aliases to gently tell you how to migrate without just erroring 🙂

[round portal]

I guess it makes sense. I'll have to rename my Movement to Move then😅

[inner sky]

this is just naming in the examples, don't worry

You don't have to rename Movement in your own code 🙂

[quasi epoch]

oh was that me

[inner sky]

happens 🙂

[quasi epoch]

what's the difference out of curiosity? why run the method multiple times?

[sharp warren]

I reset the timer after storing the variable, but before the check

[quasi epoch]

ah

[outer wyvern]

@sharp warren how are we feeling about upstreaming for the 0.18 cycle?

Are we stable enough?

[sharp warren]

I'm open to it!

[sharp warren]

Yes, I can finally say that I'm satisfied with the API.

[outer wyvern]

@here, as a non-representative users sample, how are you feeling about this?

[daring latch]

surprise ping

I don't even know the state of things

😔

[elder shore]

the 4th answer: idk

[inner sky]

100% yes from my side

[glad hawk]

very surprising but I think having to get a 3rd party solution for proper input handling does feel weird

[daring latch]

I thought there was a competing crate for input handling

[tender cave]

@summer patrol

[valid harbor]

If I had to reccomend bevy to someone, I would 100% tell them to handle inputs with a library and to use BEI specifically, so yes.

[outer wyvern]

We're running into problems with the lack of this in terms of examples / UI / editor

So I'm inclined to push this forward

[outer wyvern]

Yes; #1034547742262951966, which is my baby

[valid harbor]

leafwing input manager was originally the one on track to be upstreamed, but BEI is slightly better

[daring latch]

Oh, I didn't realize we weren't in that channel x)

[outer wyvern]

Yeah, observers threw a total wrench in my design 😂

[daring latch]

then I know the state of things even less

[cinder grail]

Those darned observers!

[inner sky]

The only major issue I have with BEIs API is how quickly it breaks down when you need input accumulation between fixed updates

which happens very quickly in 3D

But that's more of a Bevy issue than a BEI issue tbh

[slim field]

i like BEI but it had a major api refactor very recently so it doesn't seem clear to me that it's fully baked yet

[outer wyvern]

Yeah, I think that this is largely orthogonal to BEI

[cinder grail]

I think that the biggest risk factor is that we may later come up with a better architecture, but there's no way to predict when, or even if, this might happen.

[slim field]

I'm also not super fond of the macro API but perhaps there are good reasons for it; I don't know what went into the design

[outer wyvern]

My understanding is that this was mirroring the children! macro. I can try and add docs demonstrating alternatives?

[sharp warren]

That's exactly what happened with LWIM.
Observers simply didn't exist back then.

I really hoped for BSN for 0.17.
With bsn bindings would become a scene.

[cinder grail]

Yep. You don't know what the next idea is going to be until it comes; and a lot of innovation happens, not because someone had a brainstorm, but because the underlying landscape changed.

[sharp warren]

Until then, we either have to use a macro or functional API (spawn_related).

[cinder grail]

To quote Bruce Sterling, "When the world was given the gift of TCP/IP, all of a sudden a billion ideas became obvious."

[brittle swallow]

In terms of API stability, it seems like a bit of a slippery slope. Do we wait until BSN has stabilized so BEI can integrate it? But it'll be another API shift, so do we wait another cycle for it to settle? Or do we accept that it may need to change upstream and just pull the trigger.

[cinder grail]

We are still going to need solutions for people who don't want to use BSN, for whatever reason.

[sharp warren]

Yep, I liked the approach, but I wasn’t fully satisfied.
Then this new idea got stuck in my head 😅
I went through about 5 iterations of the crate or so.
But now I can say: I'm satisfied. Unless Bevy introduces some mega-observers that are fundamentally different or something like that 😅

[brittle swallow]

On the other hand, if the upstreaming calcifies the API, it does feel a little scary with BSN being imminent. I think BSN will have a profound impact on the shape of the ecosystem.

[slim field]

As an example of why I dislike the macro API, I had a bug the other day where this was crashing my game and it took me a while to figure out why

        commands.entity(entity).insert(actions![
            MyContext[(
                (
                    Action::<CancelAction>::new(),
                    ActionSettings {
                        require_reset: true,
                        ..default()
                    },
                    bindings![KeyCode::Escape, GamepadButton::East, MouseButton::Right]
                ),
                (
                    Action::<DoAction>::new(),
                    ActionSettings {
                        require_reset: true,
                        ..default()
                    },
                    bindings![KeyCode::KeyE, MouseButton::Left, GamepadButton::South]
                ),
            )]
        ]);

[sharp warren]

It should be compatible already. The cool thing about the current design is it's just a hierarchy now.

[inner sky]

Do we?

[outer wyvern]

Sometimes when writing complex generic or conditional code the macro approach fails

[brittle swallow]

Well, also like @sharp warren said, it's just a hierarchy of entities. You don't need BSN to express that. It's just nicer.

[outer wyvern]

And the macro also needs to generate something

[inner sky]

yeah but then I would just look at what the macro expands to and use that

[outer wyvern]

Yeah, totally

[inner sky]

So I'm still using BSN, just not the macro

[outer wyvern]

I think it's fine to have bsn! be the standard definition

[slim field]

With all that said, maybe there is something to be said for done is better than perfect. BEI is far better than what is built into bevy today

[brittle swallow]

by a lot haha

[cinder grail]

We should probably discuss this elsewhere, but my assumption is that bsn will be a choice, just like everything else in Bevy is a choice - you don't have to use bsn, any more than you have to use bevy's renderer.

[slim field]

actually i change my vote, i'm pro upstreaming

[sharp warren]

Ah, an extra brace? Yeah, it's annoying.
But it’s the same problem with any ECS hierarchy. It's a broader inconvenience that needs to be solved on the Bevy side.
And it most likely will be with BSN.

[slim field]

and i will just say again that having contextual input that works out of the box without any fiddling is absolutely awesome. one thing i thought about the other day was that it might be interesting to have more sophisticated ways of layering and combining contexts. right now it's still manual enabling/disabling through ContextActivity which is a bit brittle if you have a lot of contexts, or multiple ways of combining them and entering/exiting them.

[mossy plover]

I got pinged.

What's happened?

[sharp warren]

But you can already layer and/or combine contexts 🤔 What do you mean?

[sharp warren]

#1297361733886677036 message

[mossy plover]

Wonderful.

What features did BEI give?

(I know I was looking at it for some reason.)

[slim field]

In practice what I mean is that it would be nice to have a context stack or similar where I could simply pop off an input context without the subsystem that does the context popping necessarily having to know where it goes next. I suppose I could implement this myself as some sort of higher level abstraction for transitioning between contexts

[sharp warren]

I think I better share the link: https://github.com/simgine/bevy_enhanced_input

[sharp warren]

I'd consider an abstraction like this out of the scope.
Similar to states stack.
But yeah, you can easily implement something like this on top.

[vague temple]

@sharp warren Do you think it ought to be possible to increase the dimensional field in enum in ActionValueDim to also consider rotations? My use case is pre-buffering a clients rotation, consuming it client-side and later on server doing the same aciton. I guessed it made sense to use inputs in that scenario.
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/action/value/enum.ActionValueDim.html - Enum mentioned

Like an Axis4d

[mossy plover]

Well, not.

I'm not asking what features it has.

I just know there was some particular reason this was found helpful (over other options).

[outer wyvern]

Observer-based API, better serde, more sophisticated context handling

[outer wyvern]

I'm fine to support this if you PR it

[mossy plover]

Wait. Is controller support one ov the things it also provides?

That may be why I'm here.

[sharp warren]

If I remember correctly, you're working on a 3rd person camera game. Do you represent your movement direction in 3 axes?

[native pendant]

you can bind an action to many things

https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/binding/enum.Binding.html#variant.GamepadAxis

[outer wyvern]

Yes

[thick kiln]

Agreed, I think it's a bit soon to upstream, I'd like to see it used more

[vague temple]

You are asking me to contribute to a crate I use daily? But muh lazyness

[outer wyvern]

[thick kiln]

On 0.17 i'm getting this:

thread 'Compute Task Pool (1)' panicked at src/action.rs:131:13:
output value should be axis 2D

when using

#[derive(Debug, InputAction)]
#[action_output(Vec2)]
pub struct Movement;

and

Bindings::spawn(Cardinal::wasd_keys()),

probably my fault, but do I have to change anything?

[fallow ivy]

Hi, I'm trying to use this crate but it doesn't seem to work, the observer are never triggered. Am I missing something?
Here's a small project for testing.

[coarse brook]

One mistake I often make is that I forget to add the context with add_input_context::<YourContextName>. I can't see any context adding in your plugin function. Overall, skimming through your code, I can't find any contexts at all, so you might want to revisit some of the examples to see how to set up a context for your input bindings.

[outer wyvern]

Oh right @sharp warren we should discuss using inventory and friends for this

[fallow ivy]

Ah, yes indeed. My brain might have hidden these from me. I'll try that.

[sharp warren]

This is weird. This panic comes from unreachable!. And I can't imagine how this could even reach here 🤔
I saw a few people faced it, but we never managed to reproduce. The problem magically disappeared.
Could you try to provide me a minimal repro?
I think it's not 0.17-specific. We didn't change anything in this release.

[sharp warren]

Definitely!
People face this problem too often 😅 The only reason we didn't use it was because I wanted to wait for Bevy to adopt it for reflection first.
Since we're not getting it for 0.17, it looks like a good time to start using inventory.

[inner sky]

Can’t you leverage the reflection registry for this?

Maybe #[reflect(InputContext)]

[sharp warren]

I don't think so 🤔
I need to register an observer and a bunch of other things for each context.

Ah, wait, I think I might can...

No, never mind. I need static typing here.

I won't be able to create observer on a reflected type.

[inner sky]

Can we check the registry for which types reflect InputContext and set those observers on Startup?

[inner sky]

Hmm too bad

[sharp warren]

It's okay, using inventory directly instead of the reflection is not that bad.

[sharp warren]

Actually, I think I know a trick 🤔
I'll test it later 😅

[thick kiln]

The easiest repro I have is to go on branch https://github.com/cBournhonesque/lightyear/tree/cb/confirmed-merge, examples/bevy_enhanced_inputs
And run

rm -rf SL && cargo run --no-default-features --features=server,netcode,udp -- server 2>&1 | tee SL
rm -rf CL1 && cargo run --no-default-features --features=client,netcode,gui,udp -- client -c 1 2>&1 | tee CL1

I did change some input-related networking stuff, so maybe it's on me

[sharp warren]

I suspect there is a dimension missmatch somewhere on your side.
But BEI shouldn't panic anyway, I should fix it.

(will take a look tomorrow, going to sleep right now)

[untold bridge]

Hello, is there a way to trigger action on double click using BEI? I just cant find the right combination of conditions and events to make it work

[sharp warren]

I think you need to write a custom modifier for something like this 🤔
You can also PR it, I think it's worth to have a built-in for it.

[untold bridge]

thanks for response 🙂 I will have a go at it

[sharp warren]

In Unreal Engine they have a Combo modifier.
It's the only modifier I didn't port. But it's what could be used for double clicks.
It's marked as "beta" in Unreal and the idea behind is a bit weird: you specify the sequence of actions and you need to list all other actions that might cancel each step which is quite verbose.
I wanted something better, but didn't do it in the end.
if you have some ideas - feel free to experiement with it.

[round portal]

sooo, there is this migration guide tip:

// New
commands.add_observer(|explode: On<Explode>| {
    info!("{} exploded!", explode.entity);
    // you can also use `EntityEvent::event_target`, but we encourage
    // using direct field access when possible, for better documentation and clarity.
    info!("{} exploded!", explode.event_target());
});

but there is no field entity on bevy_enhanced_input. At least not on Start, I did not get to others yet

[native pendant]

Are you using the 0.19.0-rc.1?

[round portal]

yep

this works:

fn pause(on: On<Start<Pause>>, mut commands: Commands) {
    commands.trigger(TogglePause {
        entity: on.event_target(),
    });
}

I mean I'm fine calling event_target(), but it just kinda confusing to recommend using field in migration guide when its not always available

[heady mauve]

Looks like it is using context as the event target since there is more than one entity for the event. This is mentioned in the news update for Bevy 0.17:

Deriving EntityEvent will set EntityEvent::event_target to a field named entity by default. In some cases (such as events that have multiple entity fields), it might make sense to use a more descriptive name.

[round portal]

Oh, right. this makes sense

[sharp warren]

We didn't went with entity because the event actually contains 2 entities: context and action.

[round portal]

I'm not against it, I was just a bit confused with the recommendation from guide use .entity field, your decision makes total sense!

[sharp warren]

Yeah, I definitely see how the guide might be confusing

[sharp warren]

Was quite busy today with other things, but I will take a look tomorrow.

[thick kiln]

probably because i set the default value to bool here: https://github.com/cBournhonesque/lightyear/blob/cb%2Fconfirmed-merge/lightyear_inputs_bei/src/input_message.rs#L87-L87
it used to work in the previous version, but maybe something changed

hm it doesn't seem to be that, it might be an entity mapping issue. Don't bother investigating, i'm pretty sure it's on me

[sharp warren]

I opened https://github.com/simgine/bevy_enhanced_input/pull/216
Maybe this will help with the investigation (or at least make it less painful).

[glad hawk]

are you using cranelift codegen?

I think I had a similar error when I tried using cranelift

[thick kiln]

I think i fixed the entity mapping issues but I still get the error, will try your branch

[thick kiln]

i'm not getting this anymore after upgrading to the latest commit (previously i was on a BEI branch)

[minor dune]

Hi I realise I'm a little late to the discussion here, but I just want to mention a couple of small pain points that I've had as a user of BEI:

• There's no documentation on how to do pull-style queries when there are multiple context entities, e.g. in local multiplayer. All of the pull-style examples assume a single entity for each action. I found the solution is to query for the actions along with their ActionOf component, and then use that to get the context entity. However this leads to my next point:
• There's no ergonomic way to query for multiple actions of a context in the same system. Because the relationship is one context to many actions, it's easy to go from an action to its related context, but not the other way around. Maybe there is a "find_related" bevy api that I am missing... It's worth noting that the observer API has a similar issue - you only get the value of one action at a time. I can see this being a common use-case, where you want to handle multiple inputs at the same time because they affect each other.
• Input modifiers don't have any documented order in which they apply (when you have multiple on the same entity). I'm not sure if they actually support stacking? But in one case I needed to apply a deadzone and then a scale. I ended up making a custom modifier instead to achieve this.

Otherwise I think it's great and support it being upstreamed 😄

[sharp warren]

Strange, I didn't change anything except replacing the panic with a warning 🤔

[sharp warren]

Thanks for the feedback!
I'll document this today (how to handle multiple actions, ordering, etc.) 🙂

[sharp warren]

Check these PRs:
https://github.com/simgine/bevy_enhanced_input/pull/217
https://github.com/simgine/bevy_enhanced_input/pull/218

[limpid lake]

Can’t wait to see this package upstreamed, the transition to components api is amazing and inspiring!

[minor dune]

Thank you I think that covers everything nicely!

[neat bloom]

hi. i was using Press with bevy 0.16 like this

(
                Action::<Alt>::new(),
                Press::new(1.0),
                bindings![KeyCode::ShiftLeft, GamepadButton::LeftTrigger],
            ),

but not now that bevy 0.17 uses press this is ambiguous. but i dont know where to import the BEI Press from

[sharp warren]

You need to import it explicitly because it conflicts with the picking:
https://docs.rs/bevy_enhanced_input/0.19.0/bevy_enhanced_input/condition/press/struct.Press.html

[neat bloom]

boom. thank you

[acoustic marsh]

I'm trying to use the pulse condition for an automatic weapon fire action, but I don't want the player to be able to click rapidly and circumvent it. Is there a way to do this with the pulse condition or combination with cooldown or something?

[kindred epoch]

also a question from me. I am aware of Binding::mouse_motion() for Vec2s. is it possible to separate out the X and Y axis for mouse motion so that it's two separate events with f32's? .. oh maybe i can just SwizzleAxis it?

ok that works.

But now new problem. The Horizontal and Vertical Event only fires one Axis at a time. Is it possible to make the events fire at the same time?

[outer wyvern]

Why not just use the 2D form and then process them both in that event?

[kindred epoch]

i think it just makes logical sense if X axis controls one entity and Y axis controls another entity they should be two different events.

but yeah I guess I can work around this by having one Entity as the "driver" and get a reference to the other entity in the component or something.

[sharp warren]

Try adding both cooldown and pulse 🤔

[sharp warren]

The Horizontal and Vertical Event only fires one Axis at a time
But wasn't this the goal when you separated axes? 🤔

[sharp warren]

That's totally valid approach.

[kindred epoch]

I am not sure that's the implied goal in my mind.

The observed behaviour is that when I moved the mouse diagonally I would only see "Horizontal" or "Vertical" event fires, where the other is suppressed.

I was hoping for they both fire at the same time.

I got the two action, both f32, bound to two different entities independent of each other, so in my mind that's a valid thing to do and they should be able to act in parallel with each other

but maybe i am misunderstanding something. Also for reference can I have the same binding for different action?
as in if i bind LMB to "Jump" and "Atk" at the same time, and i press LMB, should both fire or just one fire?

[heady mauve]

Did you set the actions to not consume each other? It could be that since you are grabbing mouse_motion it consumes the binding for one action and doesn't fire the other one.

[kindred epoch]

can u quickly remind me how to do that? I did a quick look and I don't really see how to do that from the quick examples.

https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/action/struct.ActionSettings.html#structfield.consume_input

oh it's an action setting, i will give that a shot, but this is most likely what i need, thanks!

[heady mauve]

Let me know how it worked out for you.

[kindred epoch]

yep, that seems to be the missing puzzle piece, no longer limit it to just one axis

Appreciate it!

[acoustic marsh]

It seemed to cause a longer cooldown than what was specified when combined. I planned to look into it more, but wasn't sure if I misunderstood the behavior

[sharp warren]

We might add a special option on pulse, but it would be nicer if it could be done with the cooldown condition

[muted pewter]

Is there a way to query which contexts are live in a system? Trying to come up with a clever way to combine bevy_enhanced_input with picking.

[sharp warren]

Yeah, check for ContextActivity<C> component.

[fallow ivy]

Hi, what would be the best practice for entities that react to the same actions ?
Let's say I have multiple entities that should react to wasd, only one can react at a time based on which one is selected.

Currently I just remove every components related to the actions (ContextPriority, ContextActivity) from the last selected entity and add them to the currently selected entity.

But I feel like it a bit clunky and there's probably a better way to do it.

[sharp warren]

You can create a special "controller" entity that will apply input to the desired entity based on your game conditions 🙂

We definitely need a "patterns" section

[fallow ivy]

Do you have a link to the documentation for that?

[sharp warren]

No because it's not a special feature, it's just a pattern. Instead of moving your context around, just create a separate entity that holds the context.

[outer wyvern]

https://github.com/simgine/bevy_enhanced_input/pull/223

[outer wyvern]

https://github.com/simgine/bevy_enhanced_input/pull/224

Alright, I think that's enough for today ❤️

Let me know what y'all think, and what details I got wrong 😅

[sharp warren]

Thank you! I'll take a look today 🙂

[native pendant]

how to get mouse motion only while hold a mouse button?

[sharp warren]

You need to use a Chord:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/condition/chord/struct.Chord.html

[native pendant]

how do i give preference for an action over another, because i need [MouseMotion, MouseButton::Left] and [MouseMotion, MouseButton::Left.with_mod(ModKeys::SHIFT)]

[sharp warren]

If you press Shift, it will execute the action with modifiers first.

[native pendant]

thanks

@sharp warren the ActionValue of the Chord is always zero, how do i transfer the values of the MouseMotion to the action of the Chord?

[sharp warren]

Ah, you need to assign Chord to the MouseMotion. So instead of 3 actions, use only 2.
I'll add this to the example.

[tame umbra]

how are repeated key pressed handled?

[native pendant]

you mean Pulse? https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/condition/pulse/struct.Pulse.html

[tame umbra]

The behavior I'm looking for is to trigger once when first pressed, wait x seconds, then start the pulse

maybe adding the same action twice, once for triggering right away, and once for pulsing after a certain amount of time? not sure if that would work or if it would just hold the action as active

[native pendant]

Fired has elapsed_seconds you could ignore the pulses that happen before n seconds

[tame umbra]

oh wickedge

[native pendant]

@sharp warren is there easy way to detect double clicks?

[tame umbra]

hmm I can't seem to make this work, holding shift should make ONGOING true?

but it stays false

oh wait i might be missing ActionState

[native pendant]

for actions like that it is Fired, not Ongoing

[tame umbra]

I do want ongoing

[native pendant]

i don't remember but ongoing only happens for a very few selection of modifiers

[tame umbra]

I want to be able to check if the button is being held down

instead of fired

[native pendant]

Fired is fired every frame

[tame umbra]

oh I need Hold

this gives the intended behavior

it seems to crash when I have duplicate key bindings

[native pendant]

what is the crash?

[native pendant]

i think you wrapped the bindings in one too many ()

[tame umbra]

oh wait

you are right

ty

[native pendant]

the blue () here

[tame umbra]

i just noticed

[sharp warren]

In Unreal Engine they have a Combo modifier for it.
It's the only modifier I didn't port. But it's what could be used for double clicks.
It's marked as "beta" in Unreal and the idea behind is a bit weird: you specify the sequence of actions and you need to list all other actions that might cancel each step which is quite verbose.
I wanted something better, but didn't do it in the end.
if you have some ideas - feel free to experiement with it.

[native pendant]

DoubleTap being Tap with 2 timers, the tap timer and the timer between taps

[tame umbra]

doubleclick is usually positional too right, where you have to click twice in the same spotish

[sharp warren]

Or you can create a modifier like this. But I think creating a more generic Combo modifier would be better to upstream to the crate 🤔

[tame umbra]

Is there a way to bind to mousewheel up/down?

[native pendant]

Binding::mouse_wheel()

@sharp warren i need 2 actions on MouseButton::Right, one on Hold and one on Tap, but only one of them runs

[sharp warren]

Try to disable input consumption for these actions

[tame umbra]

PartRotate is a f32 action, how could I map "Q" into -1.0 and "E" into 1.0?

I'm looking and SwizzleAxis and it's really not clicking for me

[sharp warren]

https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/preset/bidirectional/struct.Bidirectional.html

[tame umbra]

tyty

[tame umbra]

I was able to create this behavior w Pulse + this

feels a little janky though, optimally I would be able to attach it to bindings or actions

[sharp warren]

Can't pulse already work like you described?

[tame umbra]

more like keyboard typing repeat, you press and it fires, you hold for 0.5s, then it starts firing every 0.05s

similar to holding down a key while typing

also mousewheel doesn't fire for me with this, but the keypresses work

[sharp warren]

Ah, so you need different time for the first actuation? Makes sense. Feel free to PR a configuration for it.

[sharp warren]

I think this warning might be related:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/binding/enum.Binding.html#variant.MouseWheel

[tame umbra]

ty that fixes it

[tame umbra]

some weird behavior, with Bidirectional, if I change directions it resets the "elapsed_secs" on fire, but with Cardinal::wasd_keys(), it doesn't reset time_elapsed which breaks what I have here

[sharp warren]

Not sure what you mean

[tame umbra]

actually its a bit different than i described, switching between rotate directions or left/right directions does reset time_elapsed, but switching Up -> Right -> Down -> Left does not

[sharp warren]

I lack context

But you can try looking at logs to see how the values are processed

[tame umbra]

Since I use elapsed time to prevent the pulse from firing for a small period of time, elapsed reseting when toggling between left/right means the delayed auto shift resets properly, but it doesn't when moving Up -> Left for example

[sharp warren]

I mean that I can't really comment much on it because I don't see your exact setup.
But I don't think Pulse functions incorrectly.
For your use case you need to edit Pulse modifier. We just need an option to set a different initial delay.

[tame umbra]

Yeah I think I'm doing it in a jank way, I might look into the PR later but want to focus on the game for now

[sharp warren]

I mean you can just copy the modifier and adjust the behavior to your liking. You don't even need to upstream it in order to use it 🙂

I.e. you can define custom modifiers in your crate and they can be based on default modifiers.

[tame umbra]

I looked into the input modifiers and didn't understand how I could detect direction changes like that

[sharp warren]

The suggestion was about avoid using a hack that might solve your other problem.

I.e. about adding a parameter to configure the initial delay.

[tame umbra]

I will look into it tomorrow maybe

[sharp warren]

You can just open an issue, I might take a look into this option myself later.

[tame umbra]

just an initial delay wouldn't work, because I would also want it to trigger on the initial press

[sharp warren]

We have an option to trigger on initial press in Pulse, we just need to add an option to make the first delay after it configurable.

[tame umbra]

icic

[thick kiln]

For networking BEI i've been sending the ActionState/ActionValue of various Actions.
But that might actually be pretty expensive, since there could be a ton of different actions (that depend on various combinations of raw input key presses).
Maybe the easiest would be to just network the raw input key presses, and the compute the action on both the client and server based on those?
Note that this would not involve ActionMock, which would be used to set the values on the Action directly

[sharp warren]

This will work, but require replicating user bindings 🤔

[native pendant]

@sharp warren i have an action with Hold and an action with Tap on the same binding, both with consume_input: false, now the Tap is continuously firing on mouse movement

[sharp warren]

Enable tracing, this will show you how exactly input evaluates your conditions.

[native pendant]

that is not a feature, do you mean dropping the log level of bei to trace?

[sharp warren]

Yep:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/#troubleshooting

There could be a lot of things going on with all this modifiers and conditions. It's better to just enable logging and see what exactly happening.

[native pendant]

ShiftMouseRightAction is for the Chord of the CameraPitch

the tap should go to None and shouldn't reset to Ongoing right?

and that is only while there is mouse movement

[sharp warren]

I need to see your setup as well.

[native pendant]

#[derive(InputAction)]
#[action_output(Vec2)]
pub struct CameraPitch;

#[derive(InputAction)]
#[action_output(bool)]
pub struct ResetCameraPitch;

#[derive(InputAction)]
#[action_output(bool)]
struct ShiftMouseRightAction;

fn spawn_camera_pitch_action(commands: &mut Commands, camera: Entity) {
    let shift_mouse_right = commands
        .spawn((
            ChildOf(camera),
            ActionOf::<OrbitalCamera>::new(camera),
            Action::<ShiftMouseRightAction>::new(),
            ActionSettings {
                consume_input: false,
                ..Default::default()
            },
            Hold::new(0.25),
        ))
        .id();
    commands.spawn((
        ChildOf(shift_mouse_right),
        <BindingOf as Relationship>::from(shift_mouse_right),
        MouseButton::Right.with_mod_keys(ModKeys::SHIFT),
    ));

    let camera_pith = commands
        .spawn((
            ChildOf(camera),
            ActionOf::<OrbitalCamera>::new(camera),
            Action::<CameraPitch>::new(),
            Scale::splat(1. / 128.),
            Chord::new(vec![shift_mouse_right]),
        ))
        .id();
    commands.spawn((
        ChildOf(camera_pith),
        <BindingOf as Relationship>::from(camera_pith),
        Binding::mouse_motion().with_mod_keys(ModKeys::SHIFT),
    ));
}


fn spawn_reset_camera_pitch_action(commands: &mut Commands, camera: Entity) {
    let reset_camera_pitch = commands
        .spawn((
            ChildOf(camera),
            ActionOf::<OrbitalCamera>::new(camera),
            Action::<ResetCameraPitch>::new(),
            ActionSettings {
                consume_input: false,
                ..Default::default()
            },
            Tap::new(0.25),
        ))
        .id();
    commands.spawn((
        ChildOf(reset_camera_pitch),
        <BindingOf as Relationship>::from(reset_camera_pitch),
        MouseButton::Right.with_mod_keys(ModKeys::SHIFT),
    ));
}

[sharp warren]

Quick question, why you don't spawn it as a hierarchy?

[native pendant]

autism mostly, i want to have them as children of the their respective context (for actions)

i already thought up a way to deal with that though

i am being very exaggerated on how much i organize the hierarchy

[sharp warren]

Parent-child relation made for Transform 🤔

[native pendant]

i like having them as children on the inspector

[sharp warren]

Ah, it doesn't provide a way to register custom relationship types? 😢

[native pendant]

not that i know of

[sharp warren]

I see you have both Hold and Tap with 0.25 delay. I think they may trigger at the same time. Is this expected?

[sharp warren]

Also your movement uses Shift modifier. I'd expect it to consume it.

So if your mouse movement triggered, it swallows Shift

[native pendant]

The elapsed_secs on the ResetCameraPitch is continuous, it doesn't reset between each firing

[sharp warren]

Your log says that it reads true except when you release the button, in this case it reads false and the event fires.

When you hold the button, it's Ongoing.

[native pendant]

the 2 segments are not the entire log

i took just small segments to show that they go ONGOING -> ONGOING -> FIRED -> ONGOING -> ONGOING -> ONGOING -> FIRED

while holding mouse right and moving the mouse

[native pendant]

adding cosume_input: false to the Chord improves

[sharp warren]

I'm not sure what this means

The log looks expected to me. When you start holding, it becomes Ongoing. When you release it in time - it's Fired.

[native pendant]

my hypothesis is that when the Hold gets fired, it goes to the Chord and consumes the Shift which causes the Hold and Tap to reset

[native pendant]

that is the thing, i'm not releasing

[sharp warren]

Yes, if some other action consumes the input - it will cause release.

[native pendant]

i need to think of an architecture that works now, because i have also other actions that are a mirror to what i sent but without the Shift mod

[sharp warren]

I think you don't need to put shift to both chord and the action 🤔

The Shift modifier could be on either of them.

I.e. once

[native pendant]

without it being on both the action that gets triggered is the mirrored version without the mod

Right Hold + Mouse movement change camera yaw
Double Right Tap reset camera yaw
Shift Right Hold + Mouse movement change camera pitch
Shift Double Right Tap reset camera pitch

[sharp warren]

I saw in your example tap time matches the hold time. Maybe just make them different?

Actually, it doesn't matter, I think it should just work.

[sharp warren]

I see the problem now. You need to create 2 separate contexts for actions with Hold and Tap.

And assign a higher priority to the context with hold + movement.

[tame umbra]

duplicate bindings don't seem to work for me

Multiselect doesn't work

multiselect works

[sharp warren]

Actions consume input by default.

[tame umbra]

How could I prevent that?

additional ActionSettings component to the same entity?

yay that fixed it cool

[tame umbra]

I'm working on

the PR

and I managed to get it to work with something like this

this works for my rotation pulse, but for my movement pulse, because the action is constantly actuated, it never resets the held duration

I basically encounter the same problem I had earlier

I guess for my situation, i need the actuation state to "reset" whenever the value changes

i finished it

https://github.com/simgine/bevy_enhanced_input/pull/225

[sharp warren]

Thank you! Commented on it. Feel free to answer there or here.

[tame umbra]

i answered there

[weary maple]

its hacky but you could create a middleman observer, that intakes a Vec2 event and emits two events one for horizontal one for vertical.
It sucks to have an extra level of indirection but i mean

Edit: Lol didnt realise discord had kicked me into the channel that far back in the history sane and rational application developed by premium enterprise developers truly

[tame umbra]

if let chaining my beloved 😍

[tame umbra]

fixed my branch, although wasn't able to run the tests locally

[sharp warren]

Yeah, it's awesome!

Saw your changes, I'll take a proper look today after work. I think it's good for merging, but I need to update the changelog and maybe make a few stylistic changes.
Also looks like it's non-breaking, so I'll draft a new patch right after merge 🙂

[tame umbra]

morning shatur

[sharp warren]

Hi!
You mean tests fail? 🤔

[tame umbra]

tests don't run bc of some dependency issue on my computer

[sharp warren]

First time I see something like this 😅

But no problem, we can run CI.
A bit annoying that I have to confirm the run every time until I merge your first PR. I wish they add a button to autoconfirm for specific user.

[tame umbra]

yeah i'm unsure what that is

oh wait i did something silly

did a fix

[tame umbra]

last one hopefully

[native pendant]

is it a context for the hold + movement and another for the tap, or one for the ones with shift mod and one without?

[sharp warren]

Hold + movement in a separate. You want to execute it first, so if there is no mouse movement, the input won't be consumed.

Ah, and input consumption could be kept

[native pendant]

i actually got it to work putting the Tap with a higher priority

then when the tap goes to None it fallsback to the hold + movement

[tame umbra]

should be ready for merge

[outer wyvern]

https://github.com/simgine/bevy_enhanced_input/pull/224 is ready for re-review!

I think we're getting close 😅

Thanks for iterating on it with my @sharp warren

[outer wyvern]

@sharp warren round 5 or whatever is ready 😄

[sharp warren]

It's only 3rd 😅
But it looks great. Only CI needs to be fixed.

[compact island]

@sharp warren
I stumbled upon:

press::Press::default(),

and I have a question: Is there any reasoning for mod condition to not reexport all conditions?
this would allow:

condition::Press::default(),

[sharp warren]

No reason. We could re-export it for a more convenient access, yeah.
I usually just import the whole prelude 😅

[compact island]

Oh I did but Press is somewhat problematic 😆 :

[sharp warren]

Yeah, Press is a bit special, see the docs:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/condition/press/struct.Press.html

[compact island]

yeah, I know how to work around it
ok, we'll see how it turns out when upstreaming 😅
thx for answers

[outer wyvern]

CI fixed (🤞) and final suggestion to link to the presets module addressed. Thanks 😄

I do like the new structure a lot too now

[sharp warren]

I'll draft a new patch after https://github.com/simgine/bevy_enhanced_input/pull/225,
so that users can enjoy the revamped docs 🙂

[outer wyvern]

Perfect 😄

I'll be doing another docs pass next week hopefully to prep for upstreaming, but this was the bulk of the improvements needed

The docs for each item are quite good: it was just trying to explain the big picture to a new user that was hard

[tame moat]

Can i make some inputs only work when in a certain state?

I need this on the input level bcs the inputs are networked through lightyear and if the client is in the menu i need them not move with wasd and such

for example if i want rotation only working if CursorState::Focused

[sharp warren]

States are components. You can manage them with states like everything else.
If you don't want to remove/insert - you can use Disable (from Bevy) to disable the whole entity or ContextActivity<C> (provided by the crate) to disable specific context.

[mossy plover]

Can someone help me figure out why my player_jump system isn't being triggered?

I was following the simple_fly_cam.rs example for reference, to make sure I wasn't missing anything, but I seem to have everything necessary, but it just won't run.

Here is the freestanding code:

// The player component
#[derive(Component, Default, Debug)]
pub struct Player;

// The input action which represents the player jumping.
#[derive(InputAction, Debug)]
    #[action_output(bool)]
pub(crate) struct PlayerJump;

// This system works as intended, so I can confirm the player exists and there is only one.
pub(crate) fn weigh_player_down(mut transform_q:Query<&mut Transform, With<Player>>) {
  let Ok(mut transform) = transform_q.single_mut() else { return; };
  
  transform.translation.y -= 4.5;
}

// Spawns in the player.
pub(crate) fn spawn_player(mut other_players_q:Query<Entity, With<Player>>, mut cmds:Commands) {
  for player in other_players_q.iter_mut() { cmds.entity(player).despawn(); }
  
  cmds.spawn((
    actions!(Player [
      Action::<PlayerJump>::new(),
      bindings![GamepadButton::East, MouseButton::Left, KeyCode::Space]
    ]),
    
    Player,
    /* ... */
  ));
}

// This system doesn't work for some reason.
pub(crate) fn player_jump(
      trigger:On<Fire<PlayerJump>>,
  mut transform_q:Query<&mut Transform, With<Player>>
) {
  let Ok(mut transform) = transform_q.get_mut(trigger.context) else { return; };
  
  transform.translation.y += 10.;
}

and here is the code that I use to register these things:

app.add_input_context::<Player>();
app.add_observer(player_jump);

Any pointers in the right direction would help, as I haven't used bevy_enhanced_input since version 0.11, which seemingly had a very different structure to how it does now.

[mossy plover]

Wouldn't attaching a system to the input then running that system with a .run_if condition work?

[tame moat]

not sure where to, i mentioned i cant do it in the observer since it would get sent to the server anyway

i solved it with the ContextActivity

[mossy plover]

You use .run_if in the .add_systems call you use to add the system.

I am not sure if this works for observers, but I am sure it is likely you reach your observer through a system at some point.

[mossy plover]

I found the error.

It seems to have to do with the actions! macro being kinda broken.

I had:

actions!(Player [
  Action::<PlayerJump>::new(),
  bindings![GamepadButton::East, MouseButton::Left, KeyCode::Space]
])

but needed:

actions!(Player [(
  Action::<PlayerJump>::new(),
  bindings![GamepadButton::East, MouseButton::Left, KeyCode::Space]
)])

and actions! didn't tell me that the type ov (whatever is generated by) bindings! was incorrect where it would expect an Action.

Who is the creator ov bevy_enhanced_input so I can ping them?

This seems like something they should know about.

[outer wyvern]

Shatur and I, but please file an issue!

[languid thorn]

Not sure how that can be "fixed" tbh. The macro desugars the list items into a bunch of Spawn<B: Bundle>. So your first one spawns two entities. But both are valid patterns, so I don't think there's anything actionable there.

[mossy plover]

I can't.

I can't log into GitHub right now.

My phone is broken and GitHub is requiring a code from my number. 💀

[mossy plover]

Hm.

I see.

[outer wyvern]

In the context of actions specifically, we could probably enforce stronger requirements about structural validity

I really want to ship an entity inspector by default though

[languid thorn]

So like enforcing every item to be at least an Action and bindings! pair? Seems kinda hard though.

[mossy plover]

I'm not good with macros or the complex innerworkings ov a project like this. I'd be interested in seeing how it's a problem that can be navigated around.

[mossy plover]

Maybe not an Action and bindings! pair, since, apparently, there are some uses for Actions without any bindings.

At least I read that in the documentation for the macro.

There is probably something else that could be done.

[outer wyvern]

AI-controlled players is a classic

[outer wyvern]

Yeah, I don't think it would be easy

But it would probably be possible

[mossy plover]

Anything is possible with a turing-complete machine. 👍

[mossy plover]

How do you trigger unbound actions anyways?

And can you programmatically trigger actions that are keybound?

[outer wyvern]

I'd need to look it up, but you basically just inject events in the same way that the keybinding layer does

[sharp warren]

Sadly, there is nothing we can do here. That's how macro that spawn hierarchy work 😢

[sharp warren]

It will improve with BSN

[mossy plover]

What is BSN?

How will it help the situation in this regard?

(I.e. what is the current limitation that BSN eliminates?)

[mossy plover]

Interesting. Do you know which things the documentation for that would be with?

[sharp warren]

It's Bevy scene format, you can read more about it in https://discord.com/channels/691052431525675048/1264881140007702558

[sharp warren]

It will be harder to make a mistake like this.

[outer wyvern]

Mostly by virtue of better syntax

[sharp warren]

Yeah, because right now it's easy to miss a brace.

[mossy plover]

It is.

It's a very small thing.

(Though, it's a parenthesis here.)

[outer wyvern]

I would start poking at the source code for input mapping to figure out where, but ideally we would have a dedicated example for this pattern

[mossy plover]

It will be interesting to see how this will look in the future.

[sharp warren]

You can also use input mocking. We provide a convenient API for things like "mock once" and such.

[mossy plover]

TAS is another possible usecase.

[mossy plover]

Ditto question.

[sharp warren]

Ah, you want a link?

[mossy plover]

Yes, please. That would be helpful.

[sharp warren]

https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/action/struct.ActionMock.html

[sharp warren]

Came up with this idea:
https://github.com/simgine/bevy_enhanced_input/pull/226

[atomic wasp]

found a nice way to enable/disable mouse bindings while mouse is/isn't grabbed, could any problems arise from handling it like this?

fn on_grab_mouse(bindings: Query<(Entity, &Binding)>, (...)) {
    use Binding as B;
    for (id, binding) in bindings {
        if matches!( binding, B::MouseButton { .. } | B::MouseMotion { .. } | B::MouseWheel { .. }) {
            commands.entity(id)
                .insert((Disabled, DisabledByMouseGrab));
        }
    }
}

fn on_ungrab_mouse(
    bindings: Query<Entity, (With<Disabled>, With<DisabledByMouseGrab>)>,
    (...)
) {
    for id in bindings {
        commands.entity(id)
            .remove::<(Disabled, DisabledByMouseGrab)>();
    }
}

[ornate bobcat]

Just wasted 2 hours on this myself devastating glad to see an error message will be communicated going forward!

[sharp warren]

It's a totally valid way!

[ornate bobcat]

Sorry if this has been asked before (discord search sucks) but is there a typical way to re-use an existing actions logic for another action? Example: OpenInventory is a superset of PauseGame, so being able to say "this action runs this other actions logic first" or something could be helpful.

[sharp warren]

I'd probably just attach a logic to a state and just do the transition in different actions. But depends on the game.
Using a helper function could also do the trick.
Another option is to create a special event for the logic and trigger it from different actions.

[ornate bobcat]

Perfect thanks gives me plenty to chew on 🙂

[sharp warren]

The combo modifier is up:
https://github.com/simgine/bevy_enhanced_input/pull/229

[outer wyvern]

https://github.com/simgine/bevy_enhanced_input/pull/230

[outer wyvern]

oh sick!

[sharp warren]

Great, I'll take a look soon, I'm going away from computer right now 🙂

[mossy plover]

What is the 'proper' way to remove Actions, Bindings, et cetera, from an entity – such as when I want to disable the player's ability to move?

[outer wyvern]

I would add Disabled to those actions / bindings

[mossy plover]

I shouldn't remove them myself? (I don't plan on reenabling the controls.)

How do I add Disabled to them?

[outer wyvern]

Oh just despawn them then

[outer wyvern]

just insert the Disabled component to the correct entity

[mossy plover]

The player, you mean? Or how would I despawn all the related components?

I want to disable the movement and then visually show the player dropping to the bottom ov the screen, so I have to keep them around.

[outer wyvern]

No, despawn the action / binding entities themselves 🙂

Those aren't actually components: you're creating a pseudo-hierarchy of entities

I wrote some nice new crate docs the other day that might be helpful to get a better handle on the general patterns / model: https://github.com/simgine/bevy_enhanced_input/blob/master/src/lib.rs

[mossy plover]

Are those entities despawned when the entity they were spawned for is despawned?

Or are they used globally for all similar action types for all entities?

[outer wyvern]

Yes. They're scoped to a single entity's actions

Since you need to store the state for each instance of the action seperately

[mossy plover]

I don't see Disabled in the prelude(?).

Is it part ov base bevy rather than bevy_enhanced_input?

[mossy plover]

@outer wyvern, are you referring to values ov ContextActivity? That's what Google's AI overview brought up and I wanted to make sure that was the right thing.

[outer wyvern]

Base bevy

[sharp warren]

I'd go with Alice's advice, but if you truly want to remove the context, see this:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/#removing-contexts

[sharp warren]

It's another option. If you want your player entity to be used and only disable specific context.

[sharp warren]

Once upstreamed, maybe we can improve the ergnomoics:
https://github.com/bevyengine/bevy/issues/20753
https://github.com/bevyengine/bevy/issues/20252
This would allow just .remove::<ContextName>().

[mossy plover]

I did see this and this is what I have implemented.

I'm not exactly sure how to remove the entities other than this method.

(Isn't this technically the same as what @outer wyvern suggested?)

[sharp warren]

Disabling an entity hides the entity from queries.

[sharp warren]

@outer wyvern could you add https://github.com/simgine/bevy_enhanced_input/pull/225 to your merge train? I'll draft a new patch release after it 🙂

[languid thorn]

Does this plugin have a built-in way of responding to input outside of observers? With bevy-tnua you need to move the controller in FixedUpdate and you also need to know if an action is not being held. Is there a way to do this?

[sharp warren]

Yes, you we provide a pull-style API, just query for an action state and events bitset.

[languid thorn]

Hmm, is there a way to query related ActionOf<T> entities from the Actions<T> main entity? Both Actions::entities and ActionOf::entity is private it seems.

[languid thorn]

This seem to work:

fn apply_controls(
    mut player: Query<&mut Transform, With<Actions<Player>>>,
    movement_query: Query<(&Action<Movement>, &ActionOf<Player>)>,
) -> Result {
    for (movement, action_of) in &movement_query {
        player.get_mut(action_of.get())?;
    }

    Ok(())
}

[uneven axle]

I am trying to get some observer based selection working that filters through bevy_enhanced_input to get contextpriority and such to sort it automatically but I'm struggling to figure out how to mock an input with a specific entity.

I have an observer(using bevy picking for that input) attached to my selectable entities on_click_unit that I want to fire an input for the entity that was selected. Ideally this would let me adapt to different input contexts/controls(controllers)/etc. I need the action event On<Fire<SelectUnit>> to have the entity for the original selected unit while currently it seems to be using the observer entity and I cant find a way to change that and manually provide the right entity.

pub fn on_click_unit_selection_controls(
    trigger: On<Fire<SelectUnit>>,
    actions: Single<&ContextActivity<SelectionControls>>,
    mut currently_selected_unit: ResMut<PlayerSelectedObject>,
) {
    let actions = actions.clone();
    if *actions != true {
        return;
    }
    println!("Unit selected");
    currently_selected_unit.select_unit(trigger.event().action);
}

//
pub fn on_click_unit(
    trigger: On<Pointer<Click>>,
    actions: Single<Entity, With<Action<SelectUnit>>>,
    mut commands: Commands,
) {
    println!("Unit selected");
    commands.entity(actions.entity()).insert(ActionMock::once(
        bevy_enhanced_input::action::ActionState::Fired,
        true,
    ));
}

[sharp warren]

You need to mock the action itself, not the context. Just select a child action you want to mock and do the mocking.

[uneven axle]

Do you have any examples of that?

[sharp warren]

Check the action mocking example in the docs. You just need to attach ActionMock to the action itself, not the context. This is how you select which action to mock.

[uneven axle]

I think I am doing that in the above example unless im wildly misunderstanding. I'm attaching it to the Action<> component which I copied from the ActionMock example. That action component isn't on the actual selectable objects though

[sharp warren]

I think your pointer trigger triggers on the actual entity with the context, no the action entity.

[uneven axle]

The printer trigger isn't on either the action or the context. It's on a separate entity representing the unit. It works as is without bei so I know it's on the right entity but the observers not passing the right entity to bei

[sharp warren]

I'm not sure what you mean. I see on_click_unit triggers on the selected entity and you insert ActionMock to it. But you need to insert it on the entity that represents action.

[uneven axle]

When you say represent action do you mean the entity with the Action component on it?

Or the entity that the action is happening to. Eg the unit that is being clicked?

[sharp warren]

This ^
Because from the code it looks like you insert it on the context entity.

[uneven axle]

I might be misunderstanding but I don't believe I am. on_click_unit runs first and applied the ActionMock to the entity with Action<SelectUnit> on it. The other function runs afterwards and uses the entity in the trigger which I believe is the action entity. I think my issue is that I need that entity to either be the original one that was actually selected or be able to pass that entity somehow because RN it's the middle man I believe

[sharp warren]

Wait, your units also actions?

On<Pointer<Click>> comes from picking and the associated entity is the entity with the mesh / collider. But it's also an action entity?

[uneven axle]

No they are separate entities. Im just saying I need the picking entity to be the one that's ending up in bei so I can send the correct action. The original entity seems to just be disappearing and being replaced with the action entity from bei perspective

[sharp warren]

I'm sorry, but truly don't understand you 😅
From the code you insert the mocking component on the wrong entity. You need to query for the action entity instead. Just query for the action you want to trigger using hierarchy traversal.
See the second example in the pull-style guide how to iterate over the hierarchy:
https://docs.rs/bevy_enhanced_input/latest/bevy_enhanced_input/#pull-style

[uneven axle]

Lol it's fine I think I'm confusing you or I'm really not tracking what you mean by action entity. I'm definitely inserting the MockAction onto the entity that has Action<SelectUnit> on it. It sends the action and the second observer listening to Fire<SelectUnit> activates. So I know it's all "working" from that perspective.

My issue is that I want the entity target from the original observer to be the one that is sent to the second observer and it's not. The path would ideally be (initial observer that activates through bevy_picking targeting itself -> sends that entity as the target of the action to bei -> second observer is able to send the actual game event SelectUnit with the right entity that bevy_picking started with. For some reason that second observer is reporting a different target entity and I'd like to force it to report the first entity as part of my action mock.

[sharp warren]

Ahhh, I see now! The actions.entity() confused me a little. For some reason I though you inserting ActionMock to the entity that triggered the picking event, sorry.
Not entirely sure what to suggest. I think we'll need to rework picking to make it backed by BEI after upstreaming 🤔

[uneven axle]

Rip. That would definitely be helpful because unless someone smarter than mes looked at it I'm not sure how possible/reasonable it is to use bevy picking with bei. It makes lots of stuff that seem tied to input very difficult. In my case selecting units and differentiating between whether that's a Selection vs a Move vs an Attack.

[ornate bobcat]

So I was able to mostly migrate to bei for mouse-based aim like this:

#[derive(InputAction)]
#[action_output(Vec2)]
pub(super) struct PlayerAim;

fn on_player_aim(
    player_aim: On<Fire<PlayerAim>>,
    player: Single<(&mut Player, &mut Vision, &Transform)>,
    mut window: Single<&mut Window, With<PrimaryWindow>>,
    camera: Single<(&Camera, &GlobalTransform)>,
) {
    let (mut player, mut player_vision, transform) = player.into_inner();

    let (camera, camera_transform) = *camera;

    let Ok(viewport_cursor_position) =
        camera.world_to_viewport(camera_transform, player.aim_position.extend(0.0))
    else {
        return;
    };

    player.aim_position += player_aim.value;
    window.set_cursor_position(Some(viewport_cursor_position));
}
...
// on player
(
  Action::<PlayerAim>::new(),
   Bindings::spawn((Spawn(Binding::mouse_motion()), Axial::right_stick())),
  Negate::y()
)

But I'm wondering if there is a better way because this seems both convoluted and also produces some weird results. Notably:

1. negating y works for mouse but not for the right controller stick.
2. Controller and mouse sensitives behave completely differently
3. Mouse settings like DPI do not take affect at all (duh im overriding cursor pos)
4. The Fire<MouseMotion> stops but the cursor continues to glide causing a discrepency between aim position and the on screen cursor location

Should I be doing something entirely different like hiding the cursor and using a custom sprite set to aim position?

[uneven axle]

I got it working but I dont think the solution would be recommended lol. I created a new ActionValue that holds an entity and I'm storing it in there as a Vec2 in bits form. So far it seems to work fine but I wouldn't be surprised if it breaks as it seems perilously fragile.

[ionic sundial]

A helper like this could be useful to add to BEI:

fn cardinal_with_mod_keys(
    cardinal: Cardinal<Binding, Binding, Binding, Binding>,
    mod_keys: ModKeys,
) -> Cardinal<Binding, Binding, Binding, Binding> {
    Cardinal {
        north: cardinal.north.with_mod_keys(mod_keys),
        east: cardinal.east.with_mod_keys(mod_keys),
        south: cardinal.south.with_mod_keys(mod_keys),
        west: cardinal.west.with_mod_keys(mod_keys),
    }
}

Example:

Bindings::spawn(cardinal_with_mod_keys(Cardinal::arrows(), ModKeys::CONTROL))

Not sure if there's a more ergonomic design than this being a helper function.

I couldn't find another way to add mod keys to cardinals but maybe I missed it.

[sharp warren]

I think the mouse and stick movement are inverted to each other. You probably want to invert only the mouse motion.
It's expected for different devices to have different sensitivity.
As for DPI I don't know - we just read values from Bevy.

But yeah, I don't think it's right to capture mouse movement and adjust the mouse movement at the same time.

[sharp warren]

I'd probably just do vice versa - trigger picking from BEI.

[sharp warren]

I think we could add it as a method on Cardinal

[uneven axle]

That might be smarter but I'm not sure that would do what I want. I need bevy picking to drive the actual picking and unit selection for mouse/keyboard support first of all. Secondly I wanted to use BEI for the input/action sorting. EG have a context for CombatActions, MovementActions, BaseUnitActions, UnitAbilityActions, SelectionActions, etc. Send a single event and then let bei see if it should consume it as a combat event first then go down the line if not and check movement, selection, etc.

[sharp warren]

Ah, I see

[proud coral]

What is the appropiate way to have different observers for the same action in different contexts?

For context:

I have 3 contexts: ControllingProjectile, ControllingWorld and ControllingLauncher

And i have one action called "Cylce" whcih basically allows you to cycle through things.

ControllingProjectile: It should cycle skills
ControllingWorld: It should cycle rooms
ControllingLauncher: It should cycle the loaded projectile

[heady mauve]

Are those all on the same entity?

[proud coral]

I guess I could query for the Entity with the input machine attached, and check whether it has the component that represents the context attached

[proud coral]

yes

but idk, I thought this was enough of a common case that there would be a more streamlined solution for it

[heady mauve]

Are they all active at the same time?

[proud coral]

They're not usually but the can overlap yeah

so if they overlap the observers for all the ones that are active should trigger

[heady mauve]

Is there a reason you aren't using separate CycleProjectile, CycleWorld, CycleLauncher actions? They can be bound to the same binding and if they don't consume inputs then whatever combination are active will fire.

[proud coral]

Because to me it looks like if I have the same action for different contexts

it should be the same action 🤔

if I have different actions for different contexts too

what is the point of contexts

I can have everything in the same context and just create different actions