On Sat, Jul 31, 2010 at 11:28:20AM +0200, Henrik Rydberg wrote: > On 07/30/2010 11:41 PM, Éric Piel wrote: > > [...] > > >> It is true that there are no specific section on how to handle the ABS_X/Y > > >> events, and different drivers have slightly different strategies. However, > >> common to them all is that a single finger is consistently used to report the > >> events. > > Maybe it would be useful to add a paragraph to the mt-protocol document > > about the relation with the single-touch events. > > > Indeed. > > > Well as an anecdote, on my old laptop with a single-touch synaptics > > hardware, when I press 2 fingers, the average point is reported. That's > > all in hardware, though so it probably doesn't count as a typical driver > > behaviour ;-) > > > Those devices have no other information to report. > > >>> More specifically, on this hardware, when there are two fingers, we get > > >>> only the lower coordinates and the higher coordinates, but we don't know > >>> exactly where are the fingers. As there is no tracking, the lower > >>> coordinate might be the second finger applied, so even if we report just > >>> the lower coordinates, there is 75% chance that the cursor will jump. > >> > >> > >> Perhaps it is better to not report ABS_X/Y at all in the case of two fingers on > >> this trackpad. With any reasonable userspace driver, the ABS_X/Y will be masked > >> out anyways. > > What do you mean? ABS_X/Y is still very useful, for example for gesture, > > or to manage horizontal/vertical scrolling. In the xorg synaptics > > driver, it's not used to move the cursor position, but it's definitely > > not masked out! > > > Maybe the idea sounds radical, but there is a rationale. > > The semantics of ABS_X/Y is a singular, well-defined point. In the trackpad > case, the movement of that point is used to guide a pointer on screen, but there > are other scenarios. > > In the presence of multiple touch points, both ABS_X/Y and the pointer are > ill-defined. This simple fact has far-reaching implications; in Xorg land, there > is a whole new input protocol cooking to deal with it. For all practical > purposes, the traditional ABS_X/Y point has no meaning in the multitouch case. > > Instead of extending the semantics of ABS_X/Y, one could simply omit the event > during a multi-finger touch. In practice, this would most likely confuse some > applications. Sending the last known one-finger touch point should be less > confusing. Not quite equivalent, but close enough. > > >>> That's actually the reason I prefer the average: it gives a 100% > >>> expectable behavior (because the average is always correct). In > >>> practice, when using the xorg synaptics driver, with the default > >>> 2-fingers-scrolling mode, the cursor never jumps because as soon as the > >>> second finger is applied, X and Y are not used to move the cursor. > >> > >> > >> It is predictable, but it is still bad behavior. ;-) > > : > > I'm completely against not reporting the position with 2 fingers, > > because definitely, users would lose some features. With 3 and 4 > > fingers, we report the lower coordinates (the only ones provided by the > > hardware). So it'd be extremely weird not to report any for 2 fingers! > > > This sounds like a misunderstanding. To rephrase: the suggestion is to always > send the last known one-finger touch point, and to update that point only when > there is a single finger on the pad. > This would break two-finger scrolling mode of synaptics X driver. > > > As a user, I enjoy better when the average is reported, because > > whichever finger I move, I get a feedback, there is never a finger > > "hidden" behind the other one. Nevertheless, if consistency matters, > > let's just leave it as is. > > > This behavior can be created in userland using the MT touch points, if so preferred. > -- Dmitry -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
