On my iBook, when I have gpm reading from /dev/input/mice in imps2
format, and repeating in ms3 format, scroll wheel events from my MS
"Wheel Mouse Optical USB" are *mostly* repeated correctly, except that
turning the wheel back and forth one notch seems to have no effect.

I fiddled around with strace, and discovered the following:

- A turn of the wheel 'down' will repeat as "@\0\0\1" normally, but if
  it is moved down just after moving it up it repeats as "@\0\0\0"
  instead.  (In both cases, it's read "\10\0\0\1", so it's not the
  kernel or mouse's problem.)

- A turn of the wheel 'up' will repeat as "@\0\0\17" normally, but if
  it's preceeded by anything other than another 'wheel up', it repeats
  as "@\0\0\0" instead.  (In both cases, it's read "\10\0\0\377", so
  it's not the kernel or mouse's problem.)

My first thought was that it was an attempt to "debounce" spurious
wheel movements, but since it didn't previously behave this way (and I
couldn't find anything in the changelog about such an added "feature")
I have come to doubt this hypothesis.
---- /etc/gpm.conf ---------------------------------------------------
device=/dev/input/mice
responsiveness=
repeat_type=ms3
type=imps2
append=""
sample_rate=
----------------------------------------------------------------------

This bug appears to have gone away.  I thought it might be because gpm
had been upgraded by an apt-get dist-upgrade, but I'm still using the
same version of apt, and the same configuration file.

The bug was previously quite resistant to restarting gpm,
unplugging/replugging the mouse, etc., so I don't think it's a
heisenbug.  Perhaps it was somehow fixed by an upgrade of one of the
depended-upon packages?

Anyway, unless there have been any related reports, the bug can
probably be closed.

I take back my previous assertion that the bug had mysteriously gone
away.  It turns out that xserver-xfree86 was getting input directly from
/dev/input/mice as well as from /dev/gpmdata (see bug #154057), with the
result being that it got correct scroll events from the real device.
The problem had thus been "fixed" when I upgraded xserver-common and
dexconf overwrote my XF86Conf-4 file; when I fixed my XF86Conf-4 file,
this bug reappeared.  Not mysterious at all, as it turns out.

The problem is still in existance and is exactly as I described it.