Game: Steam Input causes incorrect controller and camera behavior #1563

Open
opened 2026-11-18 07:17:56 +00:00 by BeanBot · 5 comments
Member

Reported on Discord

What happened

The reporter says Steam Input works incorrectly with their controller. It seems to operate partly in “lizard mode” before the Steam client is enabled; after Steam picks it up and Longshot starts, inputs behave erratically. Rotating the environment is difficult because they have to hold downward while walking to prevent the camera from panning upward. In lizard mode, only A, B, and the triggers appear to work; X and Y do not.

They plan to use SISR to make the Steam controller behave like an Xbox pad (translate Steam Input to XInput), and are tuning the process before collecting data. No SISR results or logs are available yet.

Expected

Controller inputs should work correctly in Longshot when using Steam Input.

Steps to reproduce

Not known yet. The reporter has not provided the Steam launch method, active Steam Input layout/configuration, or reproduction steps.

Where in the game / How often

Not known yet.

System and version

Longshot version/channel, operating system, and exact controller model are not known yet.

Logs / screenshots

No log, in-game Bug report, or screenshot provided yet.

Not known yet

Longshot version and channel; OS and exact controller model; how Steam launches Longshot and which Steam Input layout is active; exact reproduction steps, game location, and frequency; results of SISR testing; and longshot_log.txt or an in-game Bug report.

Related but different: Longshot #296 reports an Xbox 360 controller not working in gameplay while working in the F1 menu, not Steam Input's erratic mapping/camera behavior.

Reported on [Discord](https://discord.com/channels/1569849200047427645/1572230233828884592/1572230233828884592) ### What happened The reporter says Steam Input works incorrectly with their controller. It seems to operate partly in “lizard mode” before the Steam client is enabled; after Steam picks it up and Longshot starts, inputs behave erratically. Rotating the environment is difficult because they have to hold downward while walking to prevent the camera from panning upward. In lizard mode, only A, B, and the triggers appear to work; X and Y do not. They plan to use SISR to make the Steam controller behave like an Xbox pad (translate Steam Input to XInput), and are tuning the process before collecting data. No SISR results or logs are available yet. ### Expected Controller inputs should work correctly in Longshot when using Steam Input. ### Steps to reproduce Not known yet. The reporter has not provided the Steam launch method, active Steam Input layout/configuration, or reproduction steps. ### Where in the game / How often Not known yet. ### System and version Longshot version/channel, operating system, and exact controller model are not known yet. ### Logs / screenshots No log, in-game Bug report, or screenshot provided yet. ### Not known yet Longshot version and channel; OS and exact controller model; how Steam launches Longshot and which Steam Input layout is active; exact reproduction steps, game location, and frequency; results of SISR testing; and `longshot_log.txt` or an in-game Bug report. Related but different: Longshot #296 reports an Xbox 360 controller not working in gameplay while working in the F1 menu, not Steam Input's erratic mapping/camera behavior.
Author
Member

From Discord:

The reporter describes a Steam Controller where pressing X both swings Link's sword and opens the Steam keypad. They report that Steam/controller commands are executed alongside Longshot inputs. No version, OS, launch method/layout, repro location/frequency, or log provided.

From [Discord](https://discord.com/channels/1569849200047427645/1570784344241340417/1572749856676642980): The reporter describes a Steam Controller where pressing X both swings Link's sword and opens the Steam keypad. They report that Steam/controller commands are executed alongside Longshot inputs. No version, OS, launch method/layout, repro location/frequency, or log provided.

@BeanIssues im using windows 11. i am launching longshot directly from the .exe, not using it through Steam at all.

@BeanIssues im using windows 11. i am launching longshot directly from the .exe, not using it through Steam at all.

I can attest to having the same kind of issues with a steam controller (2026 version) running on KDE Arch (Linux Kernel 7.3.4) with AMD hardware (7800x3d/9070xt).

I am running RC8 of longshot and RC11 of the launcher.

Issues are in the very first area of the game: Kokiri Forest in the main area.

I can attest to having the same kind of issues with a steam controller (2026 version) running on KDE Arch (Linux Kernel 7.3.4) with AMD hardware (7800x3d/9070xt). I am running RC8 of longshot and RC11 of the launcher. Issues are in the very first area of the game: Kokiri Forest in the main area.
Member

Claimed: this is my own open PR #152 (WIP, omgitsgela/Longshot, branch restore-keyboard-policy-longshot) — a rebuild of upstream Hookshot PR #2530's shared Keyboard Auto/On/Off policy, which fixes exactly these reports: Steam Input's Desktop layout delivers keyboard events alongside the native pad, so walking moves the camera and a pad action also reaches menus as Escape/Enter.

  • Found (Steam Deck LCD, desktop mode, 2026-11-18, owner-confirmed pattern): launching the game outside Steam (desktop icon, desktop-mode shortcut, or any direct launch) makes Steam apply its Desktop layout to the focused window: the Deck's left stick emits arrow-key presses, View emits Tab, Menu emits Escape, A emits Return, while the game also reads the SDL pad. host/nn_hid.c reads both: the movement keys tilt the left stick (move_keys_stick, HOS_KEY_MOVE_*) and the camera keys feed the right stick (HOS_KEY_CAM_*, nn_hid.c:1128-1134) — one physical stick, two inputs, camera and movement together. Game Mode and the registered Steam shortcut never hit this: Steam uses a per-app gamepad layout there, so nothing is injected. The symptom is the launch route, not a pad-driver regression.
  • Repro (no Steam needed): any direct launch + any keyboard-bound defaults (WASD/arrows exist on every platform); on desktop Linux, XTest-injected arrow keys while the pad is connected reproduce the double-control identically.
  • Working on: rebasing #152's single commit onto current main (option id 54 was taken by the merged hold-toggle; API version is now 34, not 22 — both renumbered), then an in-game run on the Steam Deck desktop route with a physical controller before the PR leaves WIP. @HumanBean's 2026-11-17 note (template + Tested on my own PC) is being filled with that run.
  • Prior work: #152 (mine, open), Hookshot #2530 (closed/draft, source preserved), #613 (different: binding pages), #1252/#95 (merged menu-nav work, doesn't cover the keyboard policy), #1588 (different: Remote Play input stop). Nothing else open claims this.
  • Wanted: @jons (comment above) — with the fix you'd get Auto by default: the game ignores keyboard bindings entirely while your Steam Controller is connected, so Steam Input's injected keys stop double-firing on Windows. If you can post your exact Steam Input layout (the Desktop one, or a gamepad template?) it helps pin which events reach you.
Claimed: this is my own open PR #152 (WIP, `omgitsgela/Longshot`, branch `restore-keyboard-policy-longshot`) — a rebuild of upstream Hookshot PR #2530's shared Keyboard Auto/On/Off policy, which fixes exactly these reports: Steam Input's Desktop layout delivers keyboard events alongside the native pad, so walking moves the camera and a pad action also reaches menus as Escape/Enter. - Found (Steam Deck LCD, desktop mode, 2026-11-18, owner-confirmed pattern): launching the game outside Steam (desktop icon, desktop-mode shortcut, or any direct launch) makes Steam apply its Desktop layout to the focused window: the Deck's left stick emits arrow-key presses, View emits Tab, Menu emits Escape, A emits Return, while the game also reads the SDL pad. `host/nn_hid.c` reads both: the movement keys tilt the left stick (`move_keys_stick`, HOS_KEY_MOVE_*) and the camera keys feed the right stick (`HOS_KEY_CAM_*`, nn_hid.c:1128-1134) — one physical stick, two inputs, camera and movement together. Game Mode and the registered Steam shortcut never hit this: Steam uses a per-app gamepad layout there, so nothing is injected. The symptom is the launch route, not a pad-driver regression. - Repro (no Steam needed): any direct launch + any keyboard-bound defaults (WASD/arrows exist on every platform); on desktop Linux, XTest-injected arrow keys while the pad is connected reproduce the double-control identically. - Working on: rebasing #152's single commit onto current main (option id 54 was taken by the merged hold-toggle; API version is now 34, not 22 — both renumbered), then an in-game run on the Steam Deck desktop route with a physical controller before the PR leaves WIP. @HumanBean's 2026-11-17 note (template + *Tested on my own PC*) is being filled with that run. - Prior work: #152 (mine, open), Hookshot #2530 (closed/draft, source preserved), #613 (different: binding pages), #1252/#95 (merged menu-nav work, doesn't cover the keyboard policy), #1588 (different: Remote Play input stop). Nothing else open claims this. - Wanted: @jons (comment above) — with the fix you'd get `Auto` by default: the game ignores keyboard bindings entirely while your Steam Controller is connected, so Steam Input's injected keys stop double-firing on Windows. If you can post your exact Steam Input layout (the Desktop one, or a gamepad template?) it helps pin which events reach you.
Member

Data point for this report, reproduced today on a Steam Deck LCD (VanGogh, RADV, SteamOS, desktop mode) with Longshot launched outside Steam, which leaves Steam's Desktop layout active over the game's window. Three separate injection paths were confirmed, and they need three different answers. Only the first is fixed (PR #152).

1. Keyboard keys. The Desktop layout sends the sticks and face buttons as keys (left stick → arrows, A → Enter, B → Escape), so walking moved the camera and a controller press also reached menus as navigation. Fixed by the Keyboard Auto/On/Off row in #152 and verified in game at 7bc7f97a: the Deck's built-in pad open through SDL, state: play, walking and the camera no longer fight.

2. Mouse buttons — this is the trigger defect. The Desktop layout sends the triggers as mouse clicks, and the game's defaults bind a face button to the left mouse button and ZL to the right one (host/nn_hid.c, mouse_to[2][2] = {{HOS_SW_Y, VK_LBUTTON}, {HOS_SW_ZL, VK_RBUTTON}}). So pressing R2 delivers a left click that fires the Y action on top of the real trigger. It is intermittent within a single session because mouse buttons only count while the pointer is captured, which is why it comes and goes. The Keyboard row deliberately leaves mouse buttons alone, so this needs its own decision: either the default binds stop putting gameplay actions on mouse buttons, or the host refuses injected clicks that arrive while a controller is connected. Related: #185 asks for keyboard and mouse to be fully disabled with a controller connected.

3. Gyro. With Motion controls at their default On, and a pad whose gyro SDL opens (hid: "Steam Deck Controller" connected (SDL), with a gyroscope), the rearm guard in host/pad_sdl.c fired 102 times in ten minutes — every ~5 s, matching its 1.2 s stillness threshold plus 3 s rearm. Each re-arm turns the sensors off and on, resets the motion filter and re-learns bias, and the game re-centres the aim, which reads as the camera moving twice for one stick input. Turning Motion controls Off stops it immediately. Candidate fixes: stop re-arming after a second consecutive static detection and log once that this pad's motion is unusable, and/or do not feed motion aiming from a handheld's own IMU when the pad in use is the Deck's built-in controls.

Not related to #884: on this Deck the trigger is detected, it is the extra mouse click that misfires.

Data point for this report, reproduced today on a Steam Deck LCD (VanGogh, RADV, SteamOS, desktop mode) with Longshot launched **outside** Steam, which leaves Steam's Desktop layout active over the game's window. Three separate injection paths were confirmed, and they need three different answers. Only the first is fixed (PR #152). **1. Keyboard keys.** The Desktop layout sends the sticks and face buttons as keys (left stick → arrows, A → Enter, B → Escape), so walking moved the camera and a controller press also reached menus as navigation. Fixed by the Keyboard Auto/On/Off row in #152 and verified in game at `7bc7f97a`: the Deck's built-in pad open through SDL, `state: play`, walking and the camera no longer fight. **2. Mouse buttons — this is the trigger defect.** The Desktop layout sends the **triggers as mouse clicks**, and the game's defaults bind a face button to the left mouse button and `ZL` to the right one (`host/nn_hid.c`, `mouse_to[2][2] = {{HOS_SW_Y, VK_LBUTTON}, {HOS_SW_ZL, VK_RBUTTON}}`). So pressing R2 delivers a left click that fires the `Y` action *on top of* the real trigger. It is intermittent within a single session because mouse buttons only count while the pointer is captured, which is why it comes and goes. The Keyboard row deliberately leaves mouse buttons alone, so this needs its own decision: either the default binds stop putting gameplay actions on mouse buttons, or the host refuses injected clicks that arrive while a controller is connected. Related: #185 asks for keyboard and mouse to be fully disabled with a controller connected. **3. Gyro.** With Motion controls at their default On, and a pad whose gyro SDL opens (`hid: "Steam Deck Controller" connected (SDL), with a gyroscope`), the rearm guard in `host/pad_sdl.c` fired **102 times in ten minutes** — every ~5 s, matching its 1.2 s stillness threshold plus 3 s rearm. Each re-arm turns the sensors off and on, resets the motion filter and re-learns bias, and the game re-centres the aim, which reads as the camera moving twice for one stick input. Turning Motion controls Off stops it immediately. Candidate fixes: stop re-arming after a second consecutive static detection and log once that this pad's motion is unusable, and/or do not feed motion aiming from a handheld's own IMU when the pad in use is the Deck's built-in controls. Not related to #884: on this Deck the trigger *is* detected, it is the extra mouse click that misfires.
Sign in to join this conversation.
No milestone
No project
No assignees
4 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot#1563
No description provided.