Free aiming cursor instead of camera panning #7

Open
opened 2026-11-07 18:27:40 +00:00 by claude · 4 comments
Member

Today the crosshair is fixed at screen center, and you aim by panning the camera (±0.32 rad). That's the biggest gap versus Terraria's free mouse cursor.

Options to evaluate:

  1. A HUD cursor image driven by look deltas, re-centering the real look each step. The server raycasts from the camera through the virtual cursor and handles dig/place/attack itself.
  2. Luanti 5.12+ set_camera (see the related engine-upgrade issue)
  3. A client-side mod (CSM), though restrictions make this unlikely

Acceptance: the camera stays still while the player can target any block within reach.

Today the crosshair is fixed at screen center, and you aim by panning the camera (±0.32 rad). That's the biggest gap versus Terraria's free mouse cursor. Options to evaluate: 1. A HUD cursor image driven by look deltas, re-centering the real look each step. The server raycasts from the camera through the virtual cursor and handles dig/place/attack itself. 2. Luanti 5.12+ `set_camera` (see the related engine-upgrade issue) 3. A client-side mod (CSM), though restrictions make this unlikely Acceptance: the camera stays still while the player can target any block within reach.
claude added this to the v0.2 Core Feel milestone 2026-11-07 18:27:40 +00:00
claude added the area/camera-controls
priority
high
type
feature
labels 2026-11-07 18:27:40 +00:00
eric added this to the Strata Roadmap project 2026-11-07 18:36:27 +00:00
eric moved this to To Do in Strata Roadmap on 2026-11-07 18:37:23 +00:00
Author
Member

Needs Eric's call (tied to #8). I evaluated the three options against Luanti 5.10 and stopped short of building one, because each has a cost that changes the game's feel or its engine floor.

  1. HUD cursor driven by look deltas (5.10-compatible). The server reads yaw/pitch deltas each step, moves a HUD crosshair image, and snaps the real look back to center. Problems: the camera is the look direction in the rig, so every re-center is a visible camera twitch at server-step rate (~10–20 Hz). Digging/placing/attacking would also have to move server-side (make nodes unpointable, implement dig timing from ctrl.dig/ctrl.place, raycast through the virtual cursor). The cursor lags by a full network round trip. It's a big rewrite and it can't be judged headless; it needs a hands-on feel test.
  2. Engine 5.12+ camera API (#8). If the newer camera API can fix a side camera independent of look, the rig, the panning and much of the input latency go away. Cost: minimum engine above Debian 13's 5.10, so players on distro packages would need a newer build (Flatpak/PPA/AppImage). I haven't verified the exact API surface yet; #8 should do that first.
  3. Client-side mod. CSM restrictions (and CSMs being off by default on most servers) make this unworkable for a distributable game.

Recommendation: Do #8 first: check the 5.12+ API in a branch. If it really gives a fixed side camera, raise min_minetest_version and build the free cursor on it. Otherwise keep the current pan-to-aim (it works within REACH) and park this issue until engine support exists. Option 1 isn't worth the rewrite for a jittery result.

Labelled status/needs-decision (or I'll note it in the devlog if that label doesn't exist). Moving on to #12 and #13 in the meantime.

**Needs Eric's call (tied to #8).** I evaluated the three options against Luanti 5.10 and stopped short of building one, because each has a cost that changes the game's feel or its engine floor. 1. **HUD cursor driven by look deltas (5.10-compatible).** The server reads yaw/pitch deltas each step, moves a HUD crosshair image, and snaps the real look back to center. Problems: the camera *is* the look direction in the rig, so every re-center is a visible camera twitch at server-step rate (~10–20 Hz). Digging/placing/attacking would also have to move server-side (make nodes unpointable, implement dig timing from `ctrl.dig`/`ctrl.place`, raycast through the virtual cursor). The cursor lags by a full network round trip. It's a big rewrite and it can't be judged headless; it needs a hands-on feel test. 2. **Engine 5.12+ camera API (#8).** If the newer camera API can fix a side camera independent of look, the rig, the panning and much of the input latency go away. Cost: minimum engine above Debian 13's 5.10, so players on distro packages would need a newer build (Flatpak/PPA/AppImage). I haven't verified the exact API surface yet; #8 should do that first. 3. **Client-side mod.** CSM restrictions (and CSMs being off by default on most servers) make this unworkable for a distributable game. **Recommendation:** Do #8 first: check the 5.12+ API in a branch. If it really gives a fixed side camera, raise `min_minetest_version` and build the free cursor on it. Otherwise keep the current pan-to-aim (it works within REACH) and park this issue until engine support exists. Option 1 isn't worth the rewrite for a jittery result. Labelled `status/needs-decision` (or I'll note it in the devlog if that label doesn't exist). Moving on to #12 and #13 in the meantime.
Owner

I agree with your recommendation. Proceed with that plan.

I agree with your recommendation. Proceed with that plan.
Author
Member

Parked, following the plan Eric approved above. #8 found that set_camera only selects first-person or third-person, through at least 5.16, so it can't give a fixed side camera or a look-independent cursor. Pan-to-aim stays.

I've removed this from the v0.2 milestone so it doesn't hold up that release, and left it open. It's worth reopening for work if the engine adds a way to separate the camera from the look direction. Eric, if you'd rather it sat in a later milestone, it only takes setting one.

Related: #9 widened the pan limit slightly (0.32 → about 0.38 rad) so everything within reach stays aimable now that the camera can sit off-centre.

**Parked, following the plan Eric approved above.** #8 found that `set_camera` only selects first-person or third-person, through at least 5.16, so it can't give a fixed side camera or a look-independent cursor. Pan-to-aim stays. I've removed this from the v0.2 milestone so it doesn't hold up that release, and left it open. It's worth reopening for work if the engine adds a way to separate the camera from the look direction. Eric, if you'd rather it sat in a later milestone, it only takes setting one. Related: #9 widened the pan limit slightly (0.32 → about 0.38 rad) so everything within reach stays aimable now that the camera can sit off-centre.
claude removed this from the v0.2 Core Feel milestone 2026-11-09 06:18:54 +00:00
Author
Member

Play-test feedback from Eric on v0.2.0 (Season VI thread, post 6):

The camera rotation when aiming at blocks is a bit odd - maybe zoom back and allow the mouse to move independently from the camera angle (which can be stationary).

The second half (a cursor independent of the camera) is what this issue is parked on: the engine ties the camera to the look direction through at least 5.16.

The first half is doable now. The rotation needed to aim is atan(reach / camera distance): about 14° at the current 22 nodes. Moving the camera back to 44 nodes with a narrower FOV shows the same area but halves the rotation to about 7°, and flattens the perspective as well. The cost is a deeper shadow slab and more map blocks between the camera and the plane. I've opened a separate issue for that.

Play-test feedback from Eric on v0.2.0 ([Season VI thread, post 6](https://discourse.eom.dev/t/luanti-season-vi/820/6)): > The camera rotation when aiming at blocks is a bit odd - maybe zoom back and allow the mouse to move independently from the camera angle (which can be stationary). The second half (a cursor independent of the camera) is what this issue is parked on: the engine ties the camera to the look direction through at least 5.16. The first half is doable now. The rotation needed to aim is `atan(reach / camera distance)`: about 14° at the current 22 nodes. Moving the camera back to 44 nodes with a narrower FOV shows the same area but halves the rotation to about 7°, and flattens the perspective as well. The cost is a deeper shadow slab and more map blocks between the camera and the plane. I've opened a separate issue for that.
Sign in to join this conversation.