Bug#424743: X11 pen driver doesn't work on tabletPC Fujitsu Stylistic 2300 (fix included)
On Tue, Jun 19, 2007 at 11:51:12PM +0200, wri-ano wrote:
> about the latest relese (named "update from J Blanchard part
> 3" . at first it didn't compile. I replaced
> -----
> *x = xf86ScaleAxis(v0, 0, priv->screen_width,
> axes[0].min_value, axes[0].max_value);
> *y = xf86ScaleAxis(v1, 0, priv->screen_height,
> axes[1].min_value, axes[1].max_value);
> ----
> by
> ----
> *x = xf86ScaleAxis(v0, 0, priv->screen_width, priv->fpitMinX,
> priv->fpitMaxX);
> *y = xf86ScaleAxis(v1, 0, priv->screen_height, priv->fpitMinY,
> priv->fpitMaxY);
> ----
> it may break the screen rotation feature introduced by Justin
> in this release.
>
> Here is the behavior : the cursor wanders erratically on the
> screen whenever I approach the pen. about 1 event on 5 is
> correctly positionned although its not regular.
>
> Surprisingly, when the system has some cpu(?) load, the pen
> just works fine, and the cursor follow it.
> for example, To make it work fine, I just need to launch a "
> while (true) do true; done" on some console screen. It's
> directly related. also works with heavy X apps, or during app
> loading time.
> After some tickling with the code, It appears that the Xorg
> core function xf86ReadSerial(), used in the core of the
> module, gives faulty results when the system is quite idle.
>
> I've added some debug messages around this function, and the
> data read from the serial device is correct when the system is
> under load, but some Bytes are duplicated or NULL are added
> when the system is at low pace. Therefore, absurd values are
> read from those bytes and most coordinates are wrong.
> I didn't spot any repeated pattern or anything regular: or
> the errors seems inserted at random.
> The erratic-ness is _proportionnal_ to the load.
>
> (Adding too heavy debug messages puts loads on the system, and
> the bug disappear...)
>
> I've tried adjusting serial parameters values line 562, namely
> Vmin and Vtime, to no avail; but I'm not skilled there and it
> may still need some tweaking.
>
> The command-line program pentest.c, gives perfect results,
> load or not load. so it is not the pen/screen duo.
> Tests are done using the same stty/setserial conf
>
> even the oldest release of this file on gitweb give the same
> erratic behavior: the core of the program didn't changed.
>
> All of this was done with PROXIMITY at 0x20, so I was wrong,
> it is not the point.
> This behavior happened before sometimes, and thinking of it,
> it was always related to the load. When running, my main
> application eats a bit of cpu, that is why it does not happen
> that often in running state. I'm switching to a new one, less
> hungry, so the bug resurfaced.
>
> the tabletPC is a fujitsu 2300 with an _active_ pen. a comment
> from 2005 speaks about possible broken support for active pen
> tablets...
>
> Any pointers welcomed, especially on _X11_ serial
> configuration (Vtime, Vmin and all) or a xf86ReadSerial()
> replacement.
I am sorry, we can't help much here. There are probably less
than 5 users of this driver in the world, including the
J. Blanchard who sent the above patches. And you know more
about this software and hardware than any of us. Since nobody
really maintains this driver, you could even become its upstream
maintainer.
You should talk to the upstream or to this J Blanchard directly
(with the xorg list in CC), or at least post a bug on
bugzilla.freedesktop.org. In both cases, post your config and
log, and all what you said here.
Brice
Reply to: