在进入源码分析前，我们先来点基础知识。下面这张图画的是元素的盒式模型，这个没有兼容性问题，有问题的是元素的宽高怎么算。以宽度为例，ff中 元素宽度=content宽度，而在ie中 元素宽度=content宽度+border宽度+padding宽度。IE8中加入了box-sizzing，该css属性有两个值：border-box、content-box分别对应ie和ff中元素宽度的工作方式。
偏移量：offsetLeft、offsetTop、offsetWidth、offsetHeight
offsetLeft：包含元素的左内边框到元素的左外边框之间的像素距离。
offsetTop：包含元素的上内边框到元素的上外边框之间的相续距离。
offsetWidth：包括元素的内容区宽度、左右内边距宽度、左右边框宽度、垂直方向滚动条的宽度之和。
offsetHeight：包括元素内容区高度、左右内边距高度、左右边框高度、水平方向滚动条的高度之和。
包含元素的引用在offsetParent属性中，offsetParent属性不一定与parentNode属性相同，比如<td>的offsetParent是<table>而不是<tr>.
客户区大小：clientWidth、clientHeight
clientWidth：元素的内容区宽度+内边距宽度
clientHeight：元素的内容区高度+内边距高度
滚动大小：scrollTop、scrollLeft、scrollWidth、scrollHeight。滚动大小指的是包含滚动内容的元素大小。
scrollTop：被隐藏在内容区域上方的像素数。
scrollLeft：被隐藏在内容区域左侧的像素数。
通过设置以上两个属性可以改变元素的滚动位置。
scrollWidth：在没有滚动条情况下，元素的内容的宽度。
scrollHeight：在没有滚动条情况下，元素内容的高度。
以上基础知识，对我们分析dom-geometry模块的代码会有不少帮助。下面我们进入源码学习阶段。
dom-geometry模块封装了许多跟盒式模型相关的函数，主要涉及：content、padding、border、margin四方面。在前面的几篇文章中我们多次提到，前端js库中对dom操作的封装最终都是要用到DOM原生的API。在此模块中，最常用的原生方法就是elemet.ownerDocument.defaultView.getComputedStyle和element.getBoundingClientRect。尽管这两个方法都存在着兼容性问题，但我们都有适当的方法来解决。
getComputedStyle方法已经在dom-style模块中介绍过（ie中使用element.currentStyle其他浏览器利用原生的getComputedStyle，在webkit中对于不在正常文档流中的元素先改变display），这里简单看一下：
1 if(has("webkit")){
2 getComputedStyle = function(/*DomNode*/ node){
3 var s;
4 if(node.nodeType == 1){
5 var dv = node.ownerDocument.defaultView;
6 s = dv.getComputedStyle(node, null);
7 if(!s && node.style){
8 node.style.display = "";
9 s = dv.getComputedStyle(node, null);
10 }
11 }
12 return s || {};
13 };
14 }else if(has("ie") && (has("ie") < 9 || has("quirks"))){
15 getComputedStyle = function(node){
16 // IE (as of 7) doesn‘t expose Element like sane browsers
17 // currentStyle can be null on IE8!
18 return node.nodeType == 1 /* ELEMENT_NODE*/ && node.currentStyle ? node.currentStyle : {};
19 };
20 }else{
21 getComputedStyle = function(node){
22 return node.nodeType == 1 /* ELEMENT_NODE*/ ?
23 node.ownerDocument.defaultView.getComputedStyle(node, null) : {};
24 };
25 }
26 style.getComputedStyle = getComputedStyle;
getComputedStyle得到的某些计算后样式是带有单位的，我们要把单位去掉。这里依赖dom-style中的toPixelValue方法：
1 var toPixel;
2 if(!has("ie")){
3 toPixel = function(element, value){
4 // style values can be floats, client code may want
5 // to round for integer pixels.
6 return parseFloat(value) || 0;
7 };
8 }else{
9 toPixel = function(element, avalue){
10 if(!avalue){ return 0; }
11 // on IE7, medium is usually 4 pixels
12 if(avalue == "medium"){ return 4; }
13 // style values can be floats, client code may
14 // want to round this value for integer pixels.
15 if(avalue.slice && avalue.slice(-2) == ‘px‘){ return parseFloat(avalue); }
16 var s = element.style, rs = element.runtimeStyle, cs = element.currentStyle,
17 sLeft = s.left, rsLeft = rs.left;
18 rs.left = cs.left;
19 try{
20 // ‘avalue‘ may be incompatible with style.left, which can cause IE to throw
21 // this has been observed for border widths using "thin", "medium", "thick" constants
22 // those particular constants could be trapped by a lookup
23 // but perhaps there are more
24 s.left = avalue;
25 avalue = s.pixelLeft;
26 }catch(e){
27 avalue = 0;
28 }
29 s.left = sLeft;
30 rs.left = rsLeft;
31 return avalue;
32 };
33 }
34 style.toPixelValue = toPixel;
函数有点复杂，对于ie浏览器只要看懂这句就行：if(avalue.slice && avalue.slice(-2) == ‘px‘){ return parseFloat(avalue); }
回到dom-geometry的源码，geom.boxModel变量代表当前浏览器中对元素使用的盒式模型，默认为content-box，同时判断了ie浏览器下的情况：
var geom = {
// summary:
// This module defines the core dojo DOM geometry API.
};
// can be either:
// "border-box"
// "content-box" (default)
geom.boxModel = "content-box";
if(has("ie") /*|| has("opera")*/){
// client code may have to adjust if compatMode varies across iframes
geom.boxModel = document.compatMode == "BackCompat" ? "border-box" : "content-box";
}
接下来的几个函数比较简单、基础，通过getComputedStyle都能直接拿到相应属性：
getPadExtents()：getComputedStyle后得到paddingLeft、paddingRight、paddingTop、paddingBottom
1 geom.getPadExtents = function getPadExtents(/*DomNode*/ node, /*Object*/ computedStyle){
2
3 node = dom.byId(node);
4 var s = computedStyle || style.getComputedStyle(node), px = style.toPixelValue,
5 l = px(node, s.paddingLeft), t = px(node, s.paddingTop), r = px(node, s.paddingRight), b = px(node, s.paddingBottom);
6 return {l: l, t: t, r: r, b: b, w: l + r, h: t + b};
7 };
getBorderExtents()：getComputedStyle后得到borderLeftWidth、borderRightWidth、borderTopWidth、borderBottomWidth；同时如果border-style设置为none，border宽度为零
1 geom.getBorderExtents = function getBorderExtents(/*DomNode*/ node, /*Object*/ computedStyle){
2 node = dom.byId(node);
3 var px = style.toPixelValue, s = computedStyle || style.getComputedStyle(node),
4 l = s.borderLeftStyle != none ? px(node, s.borderLeftWidth) : 0,
5 t = s.borderTopStyle != none ? px(node, s.borderTopWidth) : 0,
6 r = s.borderRightStyle != none ? px(node, s.borderRightWidth) : 0,
7 b = s.borderBottomStyle != none ? px(node, s.borderBottomWidth) : 0;
8 return {l: l, t: t, r: r, b: b, w: l + r, h: t + b};
9 };
getPadBorderExtents()：通过上两个方法，pad+border
1 geom.getPadBorderExtents = function getPadBorderExtents(/*DomNode*/ node, /*Object*/ computedStyle){
2 node = dom.byId(node);
3 var s = computedStyle || style.getComputedStyle(node),
4 p = geom.getPadExtents(node, s),
5 b = geom.getBorderExtents(node, s);
6 return {
7 l: p.l + b.l,
8 t: p.t + b.t,
9 r: p.r + b.r,
10 b: p.b + b.b,
11 w: p.w + b.w,
12 h: p.h + b.h
13 };
14 };
getMarginExtents()：getComputedStyle后得到marginLeft、marginRight、marginTop、marginBottom
1 geom.getMarginExtents = function getMarginExtents(node, computedStyle){
2 node = dom.byId(node);
3 var s = computedStyle || style.getComputedStyle(node), px = style.toPixelValue,
4 l = px(node, s.marginLeft), t = px(node, s.marginTop), r = px(node, s.marginRight), b = px(node, s.marginBottom);
5 return {l: l, t: t, r: r, b: b, w: l + r, h: t + b};
6 };
下面的几个函数稍微有点复杂
getMarginBox（）这个方法返回一个对象
{
t: 父元素上内边框到元素上外边距的距离，
l: 父元素左内边框到元素左外边距的距离,
w: 元素左外边距到右外边距的距离,
h: 元素上外边距到下外边距的距离
}
这个函数中主要用到上文提到的偏移量，正常情况下：
t = offsetTop,
l = offsetLeft,
w = offsetWidth + marginExtents.w,
h = offsetHeight + marginExtents.h
在这个函数中有几个兼容性问题：
1、在firefox中，如果元素的overflow样子的计算值不为visible，那么offsetLeft/offsetTop得到的值是减去borderLeftStyle/borderTopStyle后的值。这应该是firefox的bug，所以我们要对此进行修复。如果getComputedStyle中能够得到left和top那就用这两个属性代替offsetLeft和offsetTop，否则计算parentNode的border宽度，手动加上这部分值
1 if(has("mozilla")){
2 // Mozilla:
3 // If offsetParent has a computed overflow != visible, the offsetLeft is decreased
4 // by the parent‘s border.
5 // We don‘t want to compute the parent‘s style, so instead we examine node‘s
6 // computed left/top which is more stable.
7 var sl = parseFloat(s.left), st = parseFloat(s.top);
8 if(!isNaN(sl) && !isNaN(st)){
9 l = sl;
10 t = st;
11 }else{
12 // If child‘s computed left/top are not parseable as a number (e.g. "auto"), we
13 // have no choice but to examine the parent‘s computed style.
14 if(p && p.style){
15 pcs = style.getComputedStyle(p);
16 if(pcs.overflow != "visible"){
17 l += pcs.borderLeftStyle != none ? px(node, pcs.borderLeftWidth) : 0;
18 t += pcs.borderTopStyle != none ? px(node, pcs.borderTopWidth) : 0;
19 }
20 }
21 }
22 }
2、在IE8和opera中情况正好相反，offsetLeft/offsetTop包含了父元素的边框，这里我们需要把他们减去
1 if(has("opera") || (has("ie") == 8 && !has("quirks"))){
2 // On Opera and IE 8, offsetLeft/Top includes the parent‘s border
3 if(p){
4 pcs = style.getComputedStyle(p);
5 l -= pcs.borderLeftStyle != none ? px(node, pcs.borderLeftWidth) : 0;
6 t -= pcs.borderTopStyle != none ? px(node, pcs.borderTopWidth) : 0;
7 }
8 }
真个函数代码如下：
1 geom.getMarginBox = function getMarginBox(/*DomNode*/ node, /*Object*/ computedStyle){
2 node = dom.byId(node);
3 var s = computedStyle || style.getComputedStyle(node), me = geom.getMarginExtents(node, s),
4 l = node.offsetLeft - me.l, t = node.offsetTop - me.t, p = node.parentNode, px = style.toPixelValue, pcs;
5 if(has("mozilla")){
6 var sl = parseFloat(s.left), st = parseFloat(s.top);
7 if(!isNaN(sl) && !isNaN(st)){
8 l = sl;
9 t = st;
10 }else{
11 if(p && p.style){
12 pcs = style.getComputedStyle(p);
13 if(pcs.overflow != "visible"){
14 l += pcs.borderLeftStyle != none ? px(node, pcs.borderLeftWidth) : 0;
15 t += pcs.borderTopStyle != none ? px(node, pcs.borderTopWidth) : 0;
16 }
17 }
18 }
19 }else if(has("opera") || (has("ie") == 8 && !has("quirks"))){
20 if(p){
21 pcs = style.getComputedStyle(p);
22 l -= pcs.borderLeftStyle != none ? px(node, pcs.borderLeftWidth) : 0;
23 t -= pcs.borderTopStyle != none ? px(node, pcs.borderTopWidth) : 0;
24 }
25 }
26 return {l: l, t: t, w: node.offsetWidth + me.w, h: node.offsetHeight + me.h};
27 };
getContentBox()函数返回如下对象：
{
l: 元素左内边距，
t: 元素上内边距，
w: 元素内容区的宽度，
h: 元素内容区的高度
}
对象中的w和h与元素的盒式模型无关。以内容区的宽高都有两套方案：clientWidth-padingWidth或者offsetWidth-paddingWidth-borderWidth，下面是函数的源码：
1 geom.getContentBox = function getContentBox(node, computedStyle){
2 // summary:
3 // Returns an object that encodes the width, height, left and top
4 // positions of the node‘s content box, irrespective of the
5 // current box model.
6 // node: DOMNode
7 // computedStyle: Object?
8 // This parameter accepts computed styles object.
9 // If this parameter is omitted, the functions will call
10 // dojo/dom-style.getComputedStyle to get one. It is a better way, calling
11 // dojo/dom-style.getComputedStyle once, and then pass the reference to this
12 // computedStyle parameter. Wherever possible, reuse the returned
13 // object of dojo/dom-style.getComputedStyle().
14
15 // clientWidth/Height are important since the automatically account for scrollbars
16 // fallback to offsetWidth/Height for special cases (see #3378)
17 node = dom.byId(node);
18 var s = computedStyle || style.getComputedStyle(node), w = node.clientWidth, h,
19 pe = geom.getPadExtents(node, s), be = geom.getBorderExtents(node, s);
20 if(!w){
21 w = node.offsetWidth;
22 h = node.offsetHeight;
23 }else{
24 h = node.clientHeight;
25 be.w = be.h = 0;
26 }
27 // On Opera, offsetLeft includes the parent‘s border
28 if(has("opera")){
29 pe.l += be.l;
30 pe.t += be.t;
31 }
32 return {l: pe.l, t: pe.t, w: w - pe.w - be.w, h: h - pe.h - be.h};
33 };
接下来有三个私有函数setBox、isButtonTag、usersBorderBox。
setBox忽略盒式模型，直接对元素样式的width、height、left、top进行设置
1 function setBox(/*DomNode*/ node, /*Number?*/ l, /*Number?*/ t, /*Number?*/ w, /*Number?*/ h, /*String?*/ u){
2 // summary:
3 // sets width/height/left/top in the current (native) box-model
4 // dimensions. Uses the unit passed in u.
5 // node:
6 // DOM Node reference. Id string not supported for performance
7 // reasons.
8 // l:
9 // left offset from parent.
10 // t:
11 // top offset from parent.
12 // w:
13 // width in current box model.
14 // h:
15 // width in current box model.
16 // u:
17 // unit measure to use for other measures. Defaults to "px".
18 u = u || "px";
19 var s = node.style;
20 if(!isNaN(l)){
21 s.left = l + u;
22 }
23 if(!isNaN(t)){
24 s.top = t + u;
25 }
26 if(w >= 0){
27 s.width = w + u;
28 }
29 if(h >= 0){
30 s.height = h + u;
31 }
32 }
isButtonTag函数用来判断元素是否是button按钮，button元素可能是直接的<button>标签，也可能是<input type="button">，所以要对着两方面进行判断
1 function isButtonTag(/*DomNode*/ node){
2 // summary:
3 // True if the node is BUTTON or INPUT.type="button".
4 return node.tagName.toLowerCase() == "button" ||
5 node.tagName.toLowerCase() == "input" && (node.getAttribute("type") || "").toLowerCase() == "button"; // boolean
6 }
usersBorderBox判断元素的盒式模型是否为border-box，三个方面：geom的boxModel是否为border-box、元素是否为table元素，元素是否为button元素
1 function usesBorderBox(/*DomNode*/ node){
2 // summary:
3 // True if the node uses border-box layout.
4
5 // We could test the computed style of node to see if a particular box
6 // has been specified, but there are details and we choose not to bother.
7
8 // TABLE and BUTTON (and INPUT type=button) are always border-box by default.
9 // If you have assigned a different box to either one via CSS then
10 // box functions will break.
11
12 return geom.boxModel == "border-box" || node.tagName.toLowerCase() == "table" || isButtonTag(node); // boolean
13 }
setContentSize方法，设置元素内容区的大小。如果元素盒式模式是border-box，则需要在参数传入的width基础上加上padding与border的宽度，否则直接设置width、height样式。
1 geom.setContentSize = function setContentSize(/*DomNode*/ node, /*Object*/ box, /*Object*/ computedStyle){
2 // summary:
3 // Sets the size of the node‘s contents, irrespective of margins,
4 // padding, or borders.
5 // node: DOMNode
6 // box: Object
7 // hash with optional "w", and "h" properties for "width", and "height"
8 // respectively. All specified properties should have numeric values in whole pixels.
9 // computedStyle: Object?
10 // This parameter accepts computed styles object.
11 // If this parameter is omitted, the functions will call
12 // dojo/dom-style.getComputedStyle to get one. It is a better way, calling
13 // dojo/dom-style.getComputedStyle once, and then pass the reference to this
14 // computedStyle parameter. Wherever possible, reuse the returned
15 // object of dojo/dom-style.getComputedStyle().
16
17 node = dom.byId(node);
18 var w = box.w, h = box.h;
19 if(usesBorderBox(node)){
20 var pb = geom.getPadBorderExtents(node, computedStyle);
21 if(w >= 0){
22 w += pb.w;
23 }
24 if(h >= 0){
25 h += pb.h;
26 }
27 }
28 setBox(node, NaN, NaN, w, h);
29 };
setMarginBox方法，设置marginBox的宽度。该方法中不去判断元素的盒式模型，width = w-padding - border -margin。通过这种方式直接设置元素的width或height属性。这里涉及的兼容性问题，主要对于低版本浏览器，所以不去分析他。
1 geom.setMarginBox = function setMarginBox(/*DomNode*/ node, /*Object*/ box, /*Object*/ computedStyle){
2 // summary:
3 // sets the size of the node‘s margin box and placement
4 // (left/top), irrespective of box model. Think of it as a
5 // passthrough to setBox that handles box-model vagaries for
6 // you.
7 // node: DOMNode
8 // box: Object
9 // hash with optional "l", "t", "w", and "h" properties for "left", "right", "width", and "height"
10 // respectively. All specified properties should have numeric values in whole pixels.
11 // computedStyle: Object?
12 // This parameter accepts computed styles object.
13 // If this parameter is omitted, the functions will call
14 // dojo/dom-style.getComputedStyle to get one. It is a better way, calling
15 // dojo/dom-style.getComputedStyle once, and then pass the reference to this
16 // computedStyle parameter. Wherever possible, reuse the returned
17 // object of dojo/dom-style.getComputedStyle().
18
19 node = dom.byId(node);
20 var s = computedStyle || style.getComputedStyle(node), w = box.w, h = box.h,
21 // Some elements have special padding, margin, and box-model settings.
22 // To use box functions you may need to set padding, margin explicitly.
23 // Controlling box-model is harder, in a pinch you might set dojo/dom-geometry.boxModel.
24 pb = usesBorderBox(node) ? nilExtents : geom.getPadBorderExtents(node, s),
25 mb = geom.getMarginExtents(node, s);
26 if(has("webkit")){
27 // on Safari (3.1.2), button nodes with no explicit size have a default margin
28 // setting an explicit size eliminates the margin.
29 // We have to swizzle the width to get correct margin reading.
30 if(isButtonTag(node)){
31 var ns = node.style;
32 if(w >= 0 && !ns.width){
33 ns.width = "4px";
34 }
35 if(h >= 0 && !ns.height){
36 ns.height = "4px";
37 }
38 }
39 }
40 if(w >= 0){
41 w = Math.max(w - pb.w - mb.w, 0);
42 }
43 if(h >= 0){
44 h = Math.max(h - pb.h - mb.h, 0);
45 }
46 setBox(node, box.l, box.t, w, h);
47 };
position()方法，主要使用node.getBoundingClientRect() ，这个方法得到left、right、top、bottom。在老版本ie下，这个方法的基准点并不是从（0,0）开始计算的，而是以（2,2）位基准点。所以ie中这个方法得到的位置信息比实际位置多了两个像素，我们要把这两个像素减掉。基准点位置偏移两个像素，所以dcument.documentElement即<html>标签的位置也不是0；所以我们可以利用document.documentElement.getBoundingClientRect().left/top得到偏移量。减去偏移量就得到了真正的位置。（偏移量问题在IE9已经修复了，而IE8标准模式是没有这个问题，所以具体获取偏移量的细节不讨论）
1 geom.position = function(/*DomNode*/ node, /*Boolean?*/ includeScroll){
2 // summary:
3 // Gets the position and size of the passed element relative to
4 // the viewport (if includeScroll==false), or relative to the
5 // document root (if includeScroll==true).
6 //
7 // description:
8 // Returns an object of the form:
9 // `{ x: 100, y: 300, w: 20, h: 15 }`.
10 // If includeScroll==true, the x and y values will include any
11 // document offsets that may affect the position relative to the
12 // viewport.
13 // Uses the border-box model (inclusive of border and padding but
14 // not margin). Does not act as a setter.
15 // node: DOMNode|String
16 // includeScroll: Boolean?
17 // returns: Object
18
19 node = dom.byId(node);
20 var db = win.body(node.ownerDocument),
21 ret = node.getBoundingClientRect();
22 ret = {x: ret.left, y: ret.top, w: ret.right - ret.left, h: ret.bottom - ret.top};
23
24 if(has("ie") < 9){
25 // On IE<9 there‘s a 2px offset that we need to adjust for, see dojo.getIeDocumentElementOffset()
26 var offset = geom.getIeDocumentElementOffset(node.ownerDocument);
27
28 // fixes the position in IE, quirks mode
29 ret.x -= offset.x + (has("quirks") ? db.clientLeft + db.offsetLeft : 0);
30 ret.y -= offset.y + (has("quirks") ? db.clientTop + db.offsetTop : 0);
31 }
32
33 // account for document scrolling
34 // if offsetParent is used, ret value already includes scroll position
35 // so we may have to actually remove that value if !includeScroll
36 if(includeScroll){
37 var scroll = geom.docScroll(node.ownerDocument);
38 ret.x += scroll.x;
39 ret.y += scroll.y;
40 }
41
42 return ret; // Object
43 };
normalizeEvent（）方法主要针对鼠标位置信息layerX/layerY、pageX/pageY做修正；前者利用ie中的offsetX/offsetY即可，后者利用clientX+documentElement/body.scrollLeft - offset和clientY+documentElement/body.scrollTop - offset，offset即是上文提到的在ie中偏移量。
geom.normalizeEvent = function(event){
// summary:
// Normalizes the geometry of a DOM event, normalizing the pageX, pageY,
// offsetX, offsetY, layerX, and layerX properties
// event: Object
if(!("layerX" in event)){
event.layerX = event.offsetX;
event.layerY = event.offsetY;
}
if(!has("dom-addeventlistener")){
// old IE version
// FIXME: scroll position query is duped from dojo/_base/html to
// avoid dependency on that entire module. Now that HTML is in
// Base, we should convert back to something similar there.
var se = event.target;
var doc = (se && se.ownerDocument) || document;
// DO NOT replace the following to use dojo/_base/window.body(), in IE, document.documentElement should be used
// here rather than document.body
var docBody = has("quirks") ? doc.body : doc.documentElement;
var offset = geom.getIeDocumentElementOffset(doc);
event.pageX = event.clientX + geom.fixIeBiDiScrollLeft(docBody.scrollLeft || 0, doc) - offset.x;
event.pageY = event.clientY + (docBody.scrollTop || 0) - offset.y;
}
};
一个周的学习研究结束，如果您觉得这篇文章对您有帮助，请不吝点击下方推荐
时间： 2024-10-11 10:37:56