在 MacOS 中实现的鼠标光标移动加速和滚轮加速在哪里
where is mouse cursor movement acceleration and scroll wheel acceleration implemented in MacOSX
鼠标光标移动加速和滚轮加速在MacOSX中实现在哪里？
在 API 级别，Core Graphics / Quartz Event Services provides the CGEvent 类型。
在应用方面，this Chrome change review, and extracted from there this comment有很多相关有趣的评论：
// Of Mice and Men
// ---------------
//
// There are three types of scroll data available on a scroll wheel CGEvent.
// Apple's documentation ([1]) is rather vague in their differences, and not
// terribly helpful in deciding which to use. This is what's really going on.
//
// First, these events behave very differently depending on whether a standard
// wheel mouse is used (one that scrolls in discrete units) or a
// trackpad/Mighty Mouse is used (which both provide continuous scrolling).
// You must check to see which was used for the event by testing the
// kCGScrollWheelEventIsContinuous field.
//
// Second, these events refer to "axes". Axis 1 is the y-axis, and axis 2 is
// the x-axis.
//
// Third, there is a concept of mouse acceleration. Scrolling the same amount
// of physical distance will give you different results logically depending on
// whether you scrolled a little at a time or in one continuous motion. Some
// fields account for this while others do not.
//
// Fourth, for trackpads there is a concept of chunkiness. When scrolling
// continuously, events can be delivered in chunks. That is to say, lots of
// scroll events with delta 0 will be delivered, and every so often an event
// with a non-zero delta will be delivered, containing the accumulated deltas
// from all the intermediate moves. [2]
//
// For notchy wheel mice (kCGScrollWheelEventIsContinuous == 0)
// ------------------------------------------------------------
//
// kCGScrollWheelEventDeltaAxis*
// This is the rawest of raw events. For each mouse notch you get a value of
// +1/-1. This does not take acceleration into account and thus is less
// useful for building UIs.
//
// kCGScrollWheelEventPointDeltaAxis*
// This is smarter. In general, for each mouse notch you get a value of
// +1/-1, but this _does_ take acceleration into account, so you will get
// larger values on longer scrolls. This field would be ideal for building
// UIs except for one nasty bug: when the shift key is pressed, this set of
// fields fails to move the value into the axis2 field (the other two types
// of data do). This wouldn't be so bad except for the fact that while the
// number of axes is used in the creation of a CGScrollWheelEvent, there is
// no way to get that information out of the event once created.
//
// kCGScrollWheelEventFixedPtDeltaAxis*
// This is a fixed value, and for each mouse notch you get a value of
// +0.1/-0.1 (but, like above, scaled appropriately for acceleration). This
// value takes acceleration into account, and in fact is identical to the
// results you get from -[NSEvent delta*]. (That is, if you linked on Tiger
// or greater; see [2] for details.)
//
// A note about continuous devices
// -------------------------------
//
// There are two devices that provide continuous scrolling events (trackpads
// and Mighty Mouses) and they behave rather differently. The Mighty Mouse
// behaves a lot like a regular mouse. There is no chunking, and the
// FixedPtDelta values are the PointDelta values multiplied by 0.1. With the
// trackpad, though, there is chunking. While the FixedPtDelta values are
// reasonable (they occur about every fifth event but have values five times
// larger than usual) the Delta values are unreasonable. They don't appear to
// accumulate properly.
//
// For continuous devices (kCGScrollWheelEventIsContinuous != 0)
// -------------------------------------------------------------
//
// kCGScrollWheelEventDeltaAxis*
// This provides values with no acceleration. With a trackpad, these values
// are chunked but each non-zero value does not appear to be cumulative.
// This seems to be a bug.
//
// kCGScrollWheelEventPointDeltaAxis*
// This provides values with acceleration. With a trackpad, these values are
// not chunked and are highly accurate.
//
// kCGScrollWheelEventFixedPtDeltaAxis*
// This provides values with acceleration. With a trackpad, these values are
// chunked but unlike Delta events are properly cumulative.
//
// Summary
// -------
//
// In general the best approach to take is: determine if the event is
// continuous. If it is not, then use the FixedPtDelta events (or just stick
// with Cocoa events). They provide both acceleration and proper horizontal
// scrolling. If the event is continuous, then doing pixel scrolling with the
// PointDelta is the way to go. In general, avoid the Delta events. They're
// the oldest (dating back to 10.4, before CGEvents were public) but they lack
// acceleration and precision, making them useful only in specific edge cases.
//
// References
// ----------
//
// [1] <http://developer.apple.com/documentation/Carbon/Reference/QuartzEventServicesRef/Reference/reference.html>
// [2] <http://developer.apple.com/releasenotes/Cocoa/AppKitOlderNotes.html>
// Scroll to the section headed "NSScrollWheel events".
//
// P.S. The "smooth scrolling" option in the system preferences is utterly
// unrelated to any of this.
在内核层面，I/O Kit is the basic driver interface, and I guess the mouse is of the HID (Human Interface Devices) family (IOHIDFamily). The XNU kernel source can be seen here but I'm not sure if that is complete because I don't find a single match when I search for IOHIDFamily. The code might be here at IOHIDFamily. Actually I found in IOHIPointing.cpp评论：
/*
* 17 July 1998 sdouglas Initial creation
* 01 April 2002 ryepez added support for scroll acceleration
*/
从代码来看，它似乎是在那里实现的。有人可以确认吗？
另见 here.
IOHIPointing中的逻辑是如何最终到达CGEvent的？这是 Core Graphics / Quartz 的全部吗？中间是什么？那是某种马赫事件吗？有关于通用架构的任何好的概述吗？
请注意，由于正在讨论在哪里实施 mouse scroll wheel acceleration in Linux / Xorg / libinput here，我想知道这个问题。
好，问题 - 而且是一个大问题。我应该指出，我没有完整的信息——正如您所发现的，其中很多都是封闭源代码。据我所知，这些是重点：
您在“系统偏好设置”中看到的通用非 Apple 鼠标的鼠标设置（跟踪速度、滚动速度等）都在 WindowServer 中处理 - 其中 CGEvents起源。在 IOHIDPointing 等中对 HID 报告进行了一些修改，但据我所知，这主要是为了保持与古怪设备的兼容性。- 更新：（参见评论中的讨论）看起来 WindowServer 可能通过在其 IORegistry 条目上设置属性将加速参数传递给内核驱动程序。
- 我相信 Apple 触摸设备（触控板、魔术鼠标）的动量滚动实际上可能至少部分地在它们各自的闭源内核驱动程序中实现，例如
AppleMultitouchTrackpadHIDEventDriver。
- HID 堆栈的较低级别用户空间端的源代码在 IOKitUser 源包中可用。这包括
IOHIDManager 等等。
- 从device到CGEvent的旅程大致是：Device ->
IOHIDDevice -> IOHIDInterface -> IOHIDEventDriver -> [IOHIDevice] -> IOHIDSystem -> IOHIDUserClient -> IOKit 内核用户通信机制（内部使用 Mach 端口） -> IOKitUser 的 HID 部分 -> WindowServer（核心图形）。
- 您可以通过
IOHIDLibUserClient. 直接连接到 IOHIDDevice 来绕过用户空间进程的此路径
IOKitUser 来源可能会更详细地回答您的一些问题。或者，如果您正在尝试做一些具体的事情，请为此打开一个新问题。
鼠标光标移动加速和滚轮加速在MacOSX中实现在哪里？
在 API 级别，Core Graphics / Quartz Event Services provides the CGEvent 类型。
在应用方面，this Chrome change review, and extracted from there this comment有很多相关有趣的评论：
// Of Mice and Men
// ---------------
//
// There are three types of scroll data available on a scroll wheel CGEvent.
// Apple's documentation ([1]) is rather vague in their differences, and not
// terribly helpful in deciding which to use. This is what's really going on.
//
// First, these events behave very differently depending on whether a standard
// wheel mouse is used (one that scrolls in discrete units) or a
// trackpad/Mighty Mouse is used (which both provide continuous scrolling).
// You must check to see which was used for the event by testing the
// kCGScrollWheelEventIsContinuous field.
//
// Second, these events refer to "axes". Axis 1 is the y-axis, and axis 2 is
// the x-axis.
//
// Third, there is a concept of mouse acceleration. Scrolling the same amount
// of physical distance will give you different results logically depending on
// whether you scrolled a little at a time or in one continuous motion. Some
// fields account for this while others do not.
//
// Fourth, for trackpads there is a concept of chunkiness. When scrolling
// continuously, events can be delivered in chunks. That is to say, lots of
// scroll events with delta 0 will be delivered, and every so often an event
// with a non-zero delta will be delivered, containing the accumulated deltas
// from all the intermediate moves. [2]
//
// For notchy wheel mice (kCGScrollWheelEventIsContinuous == 0)
// ------------------------------------------------------------
//
// kCGScrollWheelEventDeltaAxis*
// This is the rawest of raw events. For each mouse notch you get a value of
// +1/-1. This does not take acceleration into account and thus is less
// useful for building UIs.
//
// kCGScrollWheelEventPointDeltaAxis*
// This is smarter. In general, for each mouse notch you get a value of
// +1/-1, but this _does_ take acceleration into account, so you will get
// larger values on longer scrolls. This field would be ideal for building
// UIs except for one nasty bug: when the shift key is pressed, this set of
// fields fails to move the value into the axis2 field (the other two types
// of data do). This wouldn't be so bad except for the fact that while the
// number of axes is used in the creation of a CGScrollWheelEvent, there is
// no way to get that information out of the event once created.
//
// kCGScrollWheelEventFixedPtDeltaAxis*
// This is a fixed value, and for each mouse notch you get a value of
// +0.1/-0.1 (but, like above, scaled appropriately for acceleration). This
// value takes acceleration into account, and in fact is identical to the
// results you get from -[NSEvent delta*]. (That is, if you linked on Tiger
// or greater; see [2] for details.)
//
// A note about continuous devices
// -------------------------------
//
// There are two devices that provide continuous scrolling events (trackpads
// and Mighty Mouses) and they behave rather differently. The Mighty Mouse
// behaves a lot like a regular mouse. There is no chunking, and the
// FixedPtDelta values are the PointDelta values multiplied by 0.1. With the
// trackpad, though, there is chunking. While the FixedPtDelta values are
// reasonable (they occur about every fifth event but have values five times
// larger than usual) the Delta values are unreasonable. They don't appear to
// accumulate properly.
//
// For continuous devices (kCGScrollWheelEventIsContinuous != 0)
// -------------------------------------------------------------
//
// kCGScrollWheelEventDeltaAxis*
// This provides values with no acceleration. With a trackpad, these values
// are chunked but each non-zero value does not appear to be cumulative.
// This seems to be a bug.
//
// kCGScrollWheelEventPointDeltaAxis*
// This provides values with acceleration. With a trackpad, these values are
// not chunked and are highly accurate.
//
// kCGScrollWheelEventFixedPtDeltaAxis*
// This provides values with acceleration. With a trackpad, these values are
// chunked but unlike Delta events are properly cumulative.
//
// Summary
// -------
//
// In general the best approach to take is: determine if the event is
// continuous. If it is not, then use the FixedPtDelta events (or just stick
// with Cocoa events). They provide both acceleration and proper horizontal
// scrolling. If the event is continuous, then doing pixel scrolling with the
// PointDelta is the way to go. In general, avoid the Delta events. They're
// the oldest (dating back to 10.4, before CGEvents were public) but they lack
// acceleration and precision, making them useful only in specific edge cases.
//
// References
// ----------
//
// [1] <http://developer.apple.com/documentation/Carbon/Reference/QuartzEventServicesRef/Reference/reference.html>
// [2] <http://developer.apple.com/releasenotes/Cocoa/AppKitOlderNotes.html>
// Scroll to the section headed "NSScrollWheel events".
//
// P.S. The "smooth scrolling" option in the system preferences is utterly
// unrelated to any of this.
在内核层面，I/O Kit is the basic driver interface, and I guess the mouse is of the HID (Human Interface Devices) family (IOHIDFamily). The XNU kernel source can be seen here but I'm not sure if that is complete because I don't find a single match when I search for IOHIDFamily. The code might be here at IOHIDFamily. Actually I found in IOHIPointing.cpp评论：
/*
* 17 July 1998 sdouglas Initial creation
* 01 April 2002 ryepez added support for scroll acceleration
*/
从代码来看，它似乎是在那里实现的。有人可以确认吗？ 另见 here.
IOHIPointing中的逻辑是如何最终到达CGEvent的？这是 Core Graphics / Quartz 的全部吗？中间是什么？那是某种马赫事件吗？有关于通用架构的任何好的概述吗？
请注意，由于正在讨论在哪里实施 mouse scroll wheel acceleration in Linux / Xorg / libinput here，我想知道这个问题。
好，问题 - 而且是一个大问题。我应该指出，我没有完整的信息——正如您所发现的，其中很多都是封闭源代码。据我所知，这些是重点：
您在“系统偏好设置”中看到的通用非 Apple 鼠标的鼠标设置（跟踪速度、滚动速度等）都在 WindowServer 中处理 - 其中CGEvents起源。在IOHIDPointing等中对 HID 报告进行了一些修改，但据我所知，这主要是为了保持与古怪设备的兼容性。- 更新：（参见评论中的讨论）看起来 WindowServer 可能通过在其 IORegistry 条目上设置属性将加速参数传递给内核驱动程序。
- 我相信 Apple 触摸设备（触控板、魔术鼠标）的动量滚动实际上可能至少部分地在它们各自的闭源内核驱动程序中实现，例如
AppleMultitouchTrackpadHIDEventDriver。 - HID 堆栈的较低级别用户空间端的源代码在 IOKitUser 源包中可用。这包括
IOHIDManager等等。 - 从device到CGEvent的旅程大致是：Device ->
IOHIDDevice->IOHIDInterface->IOHIDEventDriver-> [IOHIDevice] ->IOHIDSystem->IOHIDUserClient-> IOKit 内核用户通信机制（内部使用 Mach 端口） -> IOKitUser 的 HID 部分 -> WindowServer（核心图形）。 - 您可以通过
IOHIDLibUserClient. 直接连接到
IOHIDDevice 来绕过用户空间进程的此路径
IOKitUser 来源可能会更详细地回答您的一些问题。或者，如果您正在尝试做一些具体的事情，请为此打开一个新问题。