Linux tablet: Issue with XInput/GHOST? #38190

New Issue

Bastien Montagne · 2014-01-13T13:01:56+01:00

Bastien Montagne commented

2014-01-13 13:01:56 +01:00

System Information
Debian testing amd64 (ATI HD5730m, radeon free driver 7.2, if that matters).

tablet: Genius G-Pen 450, with the wizardpen driver (0.8.2) and libxi6 (1.7.2).

Blender Version
Broken: current trunk (d785c64)
Worked: ???

Short description of error
Many "tablet" events have null pressure, even though the pen is down. The effects are not much noticeable, unless you disable all smoothing/simplifying of strokes in userprefs, and draw quick strokes - in these case, I get (converted) paths with zero-radius points in the middle of thick strokes (also visible as 'gaps' in 3D view GP drawing of strokes)!

Exact steps for others to reproduce the error
Attached (tmp.txt.zip) is an xtrace dump of a (very short) session of Blender, with this simple patch applied to X11's tablet event handling code:

index 3b12d91..e959014 100644
    - a/intern/ghost/intern/GHOST_SystemX11.cpp
+++ b/intern/ghost/intern/GHOST_SystemX11.cpp
@@ -1154,6 +1154,10 @@ GHOST_SystemX11::processEvent(XEvent *xe)
                                 * around tablet surface */

                          setTabletMode(this, window, data->deviceid);

 
+                               printf("%s: raw_x: %#010x, raw_y: %#010x, raw_p: %#010x\n",
+                                      __func__, data->axis_data- [x], data->axis_data- [x], data->axis_data[2]);
+                               fflush(stdout);
+

                          window->GetTabletData()->Pressure =
                                  data->axis_data[2] / ((float)m_xtablet.PressureLevels);```


In [tmp.txt.zip](https://archive.blender.org/developer/F67738/tmp.txt.zip), you can note correct tablet events handling, like eg:
```000:>:0167: Event MotionNotify(6) detail=Normal(0x00) time=0x045cf337 root=0x000002bb event=0x03600002 child=None(0x00000000) root-x=220 root-y=333 event-x=220 event-y=284 state=Mod2 same-screen=true(0x01)
000:<:0168:  8: XInputExtension-Request(131,30): QueryDeviceState device=0x0e
000:>:0168:84: Reply to QueryDeviceState: classes={class=button(0x01) #buttons=7 button states=0x00,- [x],0x00;},{class=valuator(0x02) mode=absolute, in proximity(0x01) valuator states=0x000014a9,0x000037a2,0x00000000;};
processEvent: raw_x: 0x000014a9, raw_y: 0x000037a2, raw_p: 0000000000```
…where values from `XDeviceMotionEvent.axis_data` (`processEvent` lines) match those of XInput replies (`valuator states=…`) - my device has no tilt nor wheel data, only x/y/pressure.

However, we have many lines where there are mismatches (either null values, or shifted values, like x having pressure value, and y and pressure being null in `XDeviceMotionEvent.axis_data`):
```000:<:0192:  8: XInputExtension-Request(131,30): QueryDeviceState device=0x0e
000:>:0192:84: Reply to QueryDeviceState: classes={class=button(0x01) #buttons=7 button states=0x02,- [x],0x00;},{class=valuator(0x02) mode=absolute, in proximity(0x01) valuator states=0x0000151d,0x0000378d,0x00000105;};
processEvent: raw_x: 0x00000105, raw_y: 0000000000, raw_p: 0000000000```

I don’t know how XInput is supposed to work, maybe this is normal, but at least to my newbie eyes it looks awkward!

Note I tried to use xtrace with Gimp/Inkscape, but these apps do not seem to use the same XInput (at least Gimp uses GDK3, which uses XIinput v2, xi2). Anyway, values returned by queries to the Xorg server seem always OK, so there is a glitch somewhere either in the client part of xinput, or in our own ghost code, imho.

PS: sorry for the long post!

**System Information** Debian testing amd64 (ATI HD5730m, radeon free driver 7.2, if that matters). tablet: Genius G-Pen 450, with the wizardpen driver (0.8.2) and libxi6 (1.7.2). **Blender Version** Broken: current trunk (d785c64) Worked: ??? **Short description of error** Many "tablet" events have null pressure, even though the pen is down. The effects are not much noticeable, unless you disable all smoothing/simplifying of strokes in userprefs, and draw quick strokes - in these case, I get (converted) paths with zero-radius points in the middle of thick strokes (also visible as 'gaps' in 3D view GP drawing of strokes)! **Exact steps for others to reproduce the error** Attached ([tmp.txt.zip](https://archive.blender.org/developer/F67738/tmp.txt.zip)) is an xtrace dump of a (very short) session of Blender, with this simple patch applied to X11's tablet event handling code: ```diff --git a/intern/ghost/intern/GHOST_SystemX11.cpp b/intern/ghost/intern/GHOST_SystemX11.cpp index 3b12d91..e959014 100644 - a/intern/ghost/intern/GHOST_SystemX11.cpp +++ b/intern/ghost/intern/GHOST_SystemX11.cpp @@ -1154,6 +1154,10 @@ GHOST_SystemX11::processEvent(XEvent *xe) * around tablet surface */ ``` setTabletMode(this, window, data->deviceid); ``` + printf("%s: raw_x: %#010x, raw_y: %#010x, raw_p: %#010x\n", + __func__, data->axis_data- [x], data->axis_data- [x], data->axis_data[2]); + fflush(stdout); + ``` window->GetTabletData()->Pressure = data->axis_data[2] / ((float)m_xtablet.PressureLevels);``` ``` In [tmp.txt.zip](https://archive.blender.org/developer/F67738/tmp.txt.zip), you can note correct tablet events handling, like eg: ```000:>:0167: Event MotionNotify(6) detail=Normal(0x00) time=0x045cf337 root=0x000002bb event=0x03600002 child=None(0x00000000) root-x=220 root-y=333 event-x=220 event-y=284 state=Mod2 same-screen=true(0x01) 000:<:0168: 8: XInputExtension-Request(131,30): QueryDeviceState device=0x0e 000:>:0168:84: Reply to QueryDeviceState: classes={class=button(0x01) #buttons=7 button states=0x00,- [x],0x00;},{class=valuator(0x02) mode=absolute, in proximity(0x01) valuator states=0x000014a9,0x000037a2,0x00000000;}; processEvent: raw_x: 0x000014a9, raw_y: 0x000037a2, raw_p: 0000000000``` …where values from `XDeviceMotionEvent.axis_data` (`processEvent` lines) match those of XInput replies (`valuator states=…`) - my device has no tilt nor wheel data, only x/y/pressure. However, we have many lines where there are mismatches (either null values, or shifted values, like x having pressure value, and y and pressure being null in `XDeviceMotionEvent.axis_data`): ```000:<:0192: 8: XInputExtension-Request(131,30): QueryDeviceState device=0x0e 000:>:0192:84: Reply to QueryDeviceState: classes={class=button(0x01) #buttons=7 button states=0x02,- [x],0x00;},{class=valuator(0x02) mode=absolute, in proximity(0x01) valuator states=0x0000151d,0x0000378d,0x00000105;}; processEvent: raw_x: 0x00000105, raw_y: 0000000000, raw_p: 0000000000``` I don’t know how XInput is supposed to work, maybe this is normal, but at least to my newbie eyes it looks awkward! Note I tried to use xtrace with Gimp/Inkscape, but these apps do not seem to use the same XInput (at least Gimp uses GDK3, which uses XIinput v2, xi2). Anyway, values returned by queries to the Xorg server seem always OK, so there is a glitch somewhere either in the client part of xinput, or in our own ghost code, imho. PS: sorry for the long post!

Bastien Montagne commented

2014-01-13 13:01:56 +01:00

Changed status to: 'Open'