Fix iNNEXT SNES Controller Input

ilexp commented 5 years ago

Summary

The iNNEXT SNES Controller only works partially in Duality / OpenTK. Investigate the cause and fix this.

How to reproduce

Connect the controller and run the input sample project. The issue should be apparent immediately.
Most key / axis mappings on the controller seem to be messed up, some of it stuck at -0.71f.

Workaround

The SDL2 backend seems to correctly map this input, so placing an SDL2.dll in the Duality folder fixes this in the launcher at least.
Comparison between SDL2 and OpenTK backend results. Note how SDL2 detects more axes than OpenTK:

Analysis

Investigate input mappings. They are up-to-date with the latest OpenTK version, but the SDL2 source seems to have a lot more of them.
Extend OpenTK and Duality API to retrieve more in-depth information on what's happening. Expose controller GUIDs and mapping info in both logfiles and input sample.
Investigate the full diff for duality-master vs. the most recent develop and 4.0 to cherry-pick improvements and fixes. Look into changes in Input and Platform folders.
Investigate potential underlying bugs in OpenTK input handling that result in missing two axes.

ilexp commented 5 years ago

SDL2 seems to have switched to a different controller mapping string format between 2.0.4, 2.0.5 and 2.0.6, which means the new SDL2 source input mappings cannot be used as they are. It could be worth extending the format implementation in OpenTK to match.

Progress

Renamed IUserInput.Description and IUserInputSource.Description API to Id.
Added ProductId and ProductName to the same API, which are currently implemented using the hardware GUID and gamepad mapping name.
Added ProductId and ProductName to both logs and the Input Handling sample display.
Moved from enum-based joystick axis, button and hat addressing towards index-based to avoid an API dependency on how many of them are supported / exist on a device.

ToDo

Investigate the full diff for duality-master vs. the most recent opentk/develop and opentk/4.0 to manually cherry-pick / copy-paste improvements and fixes.
- Look into changes in Input and Platform folders.
- Work in an input update branch and split up the changes over multiple semantically grouped commits.
- Don't do a full OpenTK update, as it's out of scope for this issue.
Investigate potential underlying bugs in OpenTK input handling that result in missing two axes.

ilexp commented 5 years ago

Investigating the OpenTK Joystick GUID structure and comparing old vs. new SDL2 gamepad mapping schemes, it seems like the switch was from (by example):

ffff0000000000000000504944564944
PID VID             P I D V I D

4c056802000000000000504944564944
PID VID             P I D V I D

to

03000000ffff00000000000000000000
?       PID VID

030000004c0500006802000000000000
?       PID     VID

Which means it should be possible to make the same switch in OpenTK. Need to investigate first whether that GUID is used anywhere to query OS / hardware functionality, which might require a specific format.

ilexp commented 5 years ago

Progress

Created a new duality/upstream-merge branch where opentk/develop changes from Input and Platform folders have been merged into the Duality-specific OpenTK version. So far, nothing that looks like it would cause this bug was changed, though some smaller tweaks and other input-related bugfixes made it as well.
Looked into upstream opentk/4.0 as well, but the relevant input and platform code seems to be equivalent with the exception of style changes. Skipping this.

ToDo

Check whether Duality and the InputHandling and DualStickSpaceShooter samples still work fine with the updated OpenTK version.
- Check both editor and standalone.
- Check both with and without SDL2.
- Some minor changes to the OpenTK backend may be necessary.
If they do, update the AdamsLair.OpenTK package and update the referenced version in Duality.
Investigate potential underlying bugs in OpenTK input handling that result in missing two axes.
Investigate whether an update to the controller mapping to latest SDL2 could fix the issue.

ilexp commented 5 years ago

Progress

Checked InputHandling and DualStickSpaceShooter samples with the updated OpenTK version.
- Checked both editor and standalone.
- Checked both with and without SDL2.
Fixed flipped Y direction in gamepad thumbsticks.
Updated Duality backend code to the updated OpenTK API.
No further blocking problems related to the OpenTK input update found, seems to be fine for a merge with regard to this issue.
Found an SDL-related bug where SDL versions after 2.0.5 would fail to find the proper gamepad mapping due to SDLs changed GUID format and the new GUID not being part of the OpenTK database. However, this is unrelated to any OpenTK update changes.

ToDo

Consider fixing the SDL-related bug by updating the gamepad mapping database and GUID format to the latest SDL version. Alternatively, declare it out of scope and create a new issue to fix this later on.
Update the AdamsLair.OpenTK package and update the referenced version in Duality (see feature/opentk-input-update branch), then merge back.
Investigate potential underlying bugs in OpenTK input handling that result in missing two axes.
Investigate whether an update to the controller mapping to latest SDL2 could fix the issue.

ilexp commented 5 years ago

Progress

Added support for inversion and range operators of SDL2-compatible controller mapping strings.
Switched to the SDL 2.0.6+ unified GUID format for joystick devices, fixing the mapping related SDL2 backend bug.

ToDo

Test whether the gamepad database still works properly with some gamepads out there. Volunteers appreciated!
- Download InputTest.zip
- Run Example.exe
- Select "OpenTK / GameWindowStates" from the menu.
- Check the debug texts listing connected gamepads.
- Verify your gamepad works fine.
- Collect the GUID and name of the gamepad.
Update the AdamsLair.OpenTK package and update the referenced version in Duality (see feature/opentk-input-update branch), then merge back.
Investigate potential underlying bugs in OpenTK input handling that result in missing two axes.

ilexp commented 5 years ago

Progress

Ran the InputTest linked above with a potentially different, but probably similar SNES controller from iNNEXT with and without SDL2.

Without SDL2

examples_2018-09-24_19-10-17

Controller is not mapped, no matching GUID found.
Three axes detected, one of which looks faulty and is locked to a constant negative number. The second axis works.

With SDL2

examples_2018-09-24_19-10-46

Controller is mapped, but the name is unexpected, and some buttons get the wrong mapped labels. All buttons and axes are mapped though, so it's usable.
The GUID is slightly different from the one the OpenTK implementation generated.
Five axes detected, the first four of which seem to always share the same number.

ToDo

Look into why SDL2 and OpenTK GUIDs are different for this controller.
Look into why SDL2 and OpenTK axis counts are different, and why the first OpenTK axis is messed up.
Continue to test whether the gamepad database still works properly with some gamepads out there. Volunteers appreciated!
- Download InputTest.zip
- Run Example.exe
- Select "OpenTK / GameWindowStates" from the menu.
- Check the debug texts listing connected gamepads.
- Verify your gamepad works fine.
- Collect the GUID and name of the gamepad.
Update the AdamsLair.OpenTK package and update the referenced version in Duality (see feature/opentk-input-update branch), then merge back.

SirePi commented 5 years ago

Without SDL

PS4 DualShock 4 controller 03000000-4c05-0000-cc09-000000000000; PS4 Controller Right analog's Y axis mapped as L2 trigger (mapped 0...1, 0.5 at rest) R2 trigger mapped as Right analog's Y axis (mapped -1...1) L2 trigger is seen as R2 trigger

XBox 360 wired controller 78696e70-7574-0000-0000-000000000000; XInput Controller Working ok

With SDL 2.0.8 x64

PS4 DualShock 4 controller 03000000-4c05-0000-cc09-000000000000; PS4 Controller Working "ok" (see below)

XBox 360 wired controller 03000000-5e04-0000-8e02-000000007801; XInput Controller Working "ok" (see below)

In both cases, with SDL, the Left and Right triggers start at 0.5 - once pressed they report correctly from 0 in rest position to 1 fully pressed

ilexp commented 5 years ago

Progress

Fixed generated OpenTK GUIDs on windows to match SDL2 algorithm.
Fixed invalid OpenTK joystick axes with HID usage zero overriding the first axis.
Skipping joystick devices that report having no axes, buttons or hats.
Updated the InputTest with new binaries.

ToDo

Look into why SDL2 and OpenTK axis counts are different on the iNNEXT SNES controller.
Look into why the above PS4 controller mapping is off on via OpenTK, but fine with SDL, despite reporting the same mapping.
Continue to test whether the gamepad database still works properly with some gamepads out there. Volunteers appreciated!
- Download InputTest2.zip
- Run Example.exe
- Select "OpenTK / GameWindowStates" from the menu.
- Check the debug texts listing connected gamepads.
- Verify your gamepad works fine.
- Collect the GUID and name of the gamepad.
Update the AdamsLair.OpenTK package and update the referenced version in Duality (see feature/opentk-input-update branch), then merge back.
Open an issue in the opentk repo to discuss whether and how the fixes can be introduced to the latest official OpenTK version. Due to the internal refactoring / style changes in the 4.0 branch, they may have to be applied manually.

ilexp commented 5 years ago

@SirePi Thanks! Since the PS4 Controller seems to have some problems, can you check the InputHandling sample from the latest Duality master branch version? Does it work there? If it doesn't, does it fail in the same way?

ilexp commented 5 years ago

Talked to @SirePi about the PS4 Controller issue.

With SDL: https://gfycat.com/DescriptiveEthicalCub Without SDL: https://gfycat.com/CreepyFabulousCoypu

Without SDL, the mapping is off, one of the triggers is mapped to a sticks Y axis and vice versa. Could this be a similar issue as with the iNNEXT SNES controller, such as SDL enumerating axes differently?

ToDo

Look into why SDL2 and OpenTK axis counts are different on the iNNEXT SNES controller.
Look into why the above PS4 controller mapping is off via OpenTK, but fine with SDL, despite reporting the same mapping.
Look into why there's an extra joystick / gamepad reported without SDL.
Continue to test whether the gamepad database still works properly with some gamepads out there. Volunteers appreciated!
- Download InputTest2.zip
- Run Example.exe
- Select "OpenTK / GameWindowStates" from the menu.
- Check the debug texts listing connected gamepads.
- Verify your gamepad works fine.
- Collect the GUID and name of the gamepad.
Update the AdamsLair.OpenTK package and update the referenced version in Duality (see feature/opentk-input-update branch), then merge back.
Open an issue in the opentk repo to discuss whether and how the fixes can be introduced to the latest official OpenTK version. Due to the internal refactoring / style changes in the 4.0 branch, they may have to be applied manually.

ilexp commented 5 years ago

Progress

Looked into different axis behaviors between SDL2 and OpenTK with iNNEXT SNES.

For some reason, the controller reports its Y axis first, followed by four X axes with identical data.
SDL2 resolves the situation by sorting axes by their usage, resulting in an X, X, X, X, Y axis layout for the joystick device.
OpenTK on the other hand uses a translate / remap axis approach that maps all four X axes into one and ends up with an X, Y layout.
The difference in layout causes the controller mapping to fail with OpenTK, as it is set up to use axes 3 and 4.
Changing the way OpenTK treats axis data to mirror what SDL does could resolve this issue.
Any change to this would need to be done on all platforms, if applicable.

ToDo

Look into why SDL2 and OpenTK axis counts are different on the iNNEXT SNES controller.
Look into why the above PS4 controller mapping is off via OpenTK, but fine with SDL, despite reporting the same mapping.
Look into why there's an extra joystick / gamepad reported without SDL.
Continue to test whether the gamepad database still works properly with some gamepads out there. Volunteers appreciated!
- Download InputTest2.zip
- Run Example.exe
- Select "OpenTK / GameWindowStates" from the menu.
- Check the debug texts listing connected gamepads.
- Verify your gamepad works fine.
- Collect the GUID and name of the gamepad.
Update the AdamsLair.OpenTK package and update the referenced version in Duality (see feature/opentk-input-update branch), then merge back.
Open an issue in the opentk repo to discuss whether and how the fixes can be introduced to the latest official OpenTK version. Due to the internal refactoring / style changes in the 4.0 branch, they may have to be applied manually.

ilexp commented 5 years ago

Progress

Implemented an experimental replacement for internal OpenTK input item index mapping, aiming to mirror SDL behavior.
Verified that the iNNEXT SNES controller mapping now works in OpenTK.

ToDo

Verify behavior of the iNNEXT SNES controller is equal with and without SDL backend.
Create a new test release with the above experimental fix and ping SirePi to check whether the change improved the PS4 controller situation.
Look into why the above PS4 controller mapping is off via OpenTK, but fine with SDL, despite reporting the same mapping.
Look into why there's an extra joystick / gamepad reported without SDL.
Continue to test whether the gamepad database still works properly with some gamepads out there. Volunteers appreciated!
- Download InputTest2.zip
- Run Example.exe
- Select "OpenTK / GameWindowStates" from the menu.
- Check the debug texts listing connected gamepads.
- Verify your gamepad works fine.
- Collect the GUID and name of the gamepad.
Update the AdamsLair.OpenTK package and update the referenced version in Duality (see feature/opentk-input-update branch), then merge back.
Open an issue in the opentk repo to discuss whether and how the fixes can be introduced to the latest official OpenTK version. Due to the internal refactoring / style changes in the 4.0 branch, they may have to be applied manually.

ilexp commented 5 years ago

Progress

Verified that iNNEXT SNES controller behavior is now the same with and without SDL, both working correctly.
Created a new InputTest release with the most recent fixes.

ToDo

Look into why the above PS4 controller mapping is off via OpenTK, but fine with SDL, despite reporting the same mapping.
Look into why there's an extra joystick / gamepad reported without SDL when the PS4 controller used by SirePi is connected.
Continue to test whether the gamepad database still works properly with some gamepads out there. Volunteers appreciated!
- Download InputTest3.zip
- Run Example.exe
- Select "OpenTK / GameWindowStates" from the menu.
- Check the debug texts listing connected gamepads.
- Verify your gamepad works fine.
- Collect the GUID and name of the gamepad.
Update the AdamsLair.OpenTK package and update the referenced version in Duality (see feature/opentk-input-update branch), then merge back.
Open an issue in the opentk repo to discuss whether and how the fixes can be introduced to the latest official OpenTK version. Due to the internal refactoring / style changes in the 4.0 branch, they may have to be applied manually.

ilexp commented 5 years ago

@SirePi Can you check whether there's any change to PS4 controller behavior on the most recent InputTest3.zip download?

Edit: Talked to SirePi on the chat. No changes in PS4 controller behavior after the recent fix.

SirePi commented 5 years ago

Sadly, no changes

ilexp commented 5 years ago

Progress

Updated the AdamsLair.OpenTK package and updated the referenced version in Duality (see feature/opentk-input-update branch), merged back to master.

ToDo

Move the PS4 controller mapping bug into a separate issue.
- Look into why the above PS4 controller mapping is off via OpenTK, but fine with SDL, despite reporting the same mapping.
- Look into why there's an extra joystick / gamepad reported without SDL when the PS4 controller used by SirePi is connected.
- Continue to test whether the gamepad database still works properly with some gamepads out there.
Open an issue in the opentk repo to discuss whether and how the fixes can be introduced to the latest official OpenTK version. Due to the internal refactoring / style changes in the 4.0 branch, they may have to be applied manually.

ilexp commented 5 years ago

Progress

Opened a new issue in the OpenTK repo for discussing a potential manual merge of the previous fixes.
Opened a new issue for tackling the remaining PS4 controller problems.

Closing this.

AdamsLair / duality