本次学习drawLooper.cpp中有关SkDrawLooper类的用法,并且分析了canvas draw api中的二层循环的作用。

分析

SkDrawLooper有两个子类:SkLayerDrawLooper和SkBlurDrawLooper。

先看一下drawLooper.cpp里面的例子,主要看onDraw()做什么:

virtual void onDraw(SkCanvas* canvas) SK_OVERRIDE {  
        this->init();//初始化  
  
        SkPaint  paint;  
        paint.setAntiAlias(true);//设置抗锯齿  
        paint.setTextSize(SkIntToScalar(72));//文字大小  
        paint.setLooper(fLooper);//设置SkDrawLooper  
  
        canvas->drawCircle(SkIntToScalar(50), SkIntToScalar(50),  
                           SkIntToScalar(30), paint);//画圆  
        canvas->drawRectCoords(SkIntToScalar(150), SkIntToScalar(50),  
                               SkIntToScalar(200), SkIntToScalar(100), paint);//画矩形  
  
        canvas->drawText("Looper", 6, SkIntToScalar(230), SkIntToScalar(100),  
                         paint);//画文字  
}  

在onDraw()可以看到,这个函数在固定位置绘制了一个圆、一个矩形和一个“looper”文字。而在最终跑出的结果中可以看到,绘制的这三个图形都有一个模糊阴影,并且三个图形的边缘为红白相间,中间为蓝色填充。

drawLooper

造成这样结果的始作俑者是init()函数:

void init() {  
    if (fLooper) return;  
  
    static const struct {              //匿名结构体定义了一组描述参数  
        SkColor         fColor;        //颜色  
        SkPaint::Style  fStyle;        //path style  
        SkScalar        fWidth;        //线宽  
        SkScalar        fOffset;       //blur偏移  
        SkScalar        fBlur;         //blur sigma输入参数  
    } gParams[] = {                    //gParams定义了4组不同效果                                                                   
        { SK_ColorWHITE, SkPaint::kStroke_Style, SkIntToScalar(1)*3/4, 0, 0 },    
        { SK_ColorRED, SkPaint::kStroke_Style, SkIntToScalar(4), 0, 0 },  
        { SK_ColorBLUE, SkPaint::kFill_Style, 0, 0, 0 },  
        { 0x88000000, SkPaint::kFill_Style, 0, SkIntToScalar(10), SkIntToScalar(3) }  
    };  
  
    SkLayerDrawLooper::Builder looperBuilder;//SkLayerDrawLooper的内部类  
  
    SkLayerDrawLooper::LayerInfo info;  
    info.fPaintBits = SkLayerDrawLooper::kStyle_Bit | SkLayerDrawLooper::kMaskFilter_Bit;  
    info.fColorMode = SkXfermode::kSrc_Mode;  
  
    for (size_t i = 0; i < SK_ARRAY_COUNT(gParams); i++) {  
        info.fOffset.set(gParams[i].fOffset, gParams[i].fOffset);  
        SkPaint* paint = looperBuilder.addLayer(info);  
        paint->setColor(gParams[i].fColor);  
        paint->setStyle(gParams[i].fStyle);  
        paint->setStrokeWidth(gParams[i].fWidth);  
        if (gParams[i].fBlur > 0) {  
            SkMaskFilter* mf = SkBlurMaskFilter::Create(kNormal_SkBlurStyle,  
                                     SkBlurMask::ConvertRadiusToSigma(gParams[i].fBlur));  
            paint->setMaskFilter(mf)->unref();  
        }  
    }  
    fLooper = looperBuilder.detachLooper();  
}  

看一下init()函数中的两个类:SkLayerDrawLooper::LayerInfo和SkLayerDrawLooper::Builder。

对于SkLayerDrawLooper::LayerInfo,skia的描述如下:

/** 
     *  Info for how to apply the layer's paint and offset. 
     * 
     *  fColorMode controls how we compute the final color for the layer: 
     *      The layer's paint's color is treated as the SRC 
     *      The draw's paint's color is treated as the DST 
     *      final-color = Mode(layers-color, draws-color); 
     *  Any SkXfermode::Mode will work. Two common choices are: 
     *      kSrc_Mode: to use the layer's color, ignoring the draw's 
     *      kDst_Mode: to just keep the draw's color, ignoring the layer's 
     */  
    struct SK_API LayerInfo {  
        BitFlags            fPaintBits;  
        SkXfermode::Mode    fColorMode;  
        SkVector            fOffset;  
        bool                fPostTranslate; //!< applies to fOffset  
  
        /** 
         *  Initial the LayerInfo. Defaults to settings that will draw the 
         *  layer with no changes: e.g. 
         *      fPaintBits == 0 
         *      fColorMode == kDst_Mode 
         *      fOffset == (0, 0) 
         */  
        LayerInfo();  
    };  

init()函数中定义了info的fPaintBits、fColorMode和fOffset。

再来看SkLayerDrawLooper::Builder:

class SK_API Builder {  
public:  
    Builder();  
    ~Builder();  
  
    /** 
     *  Call for each layer you want to add (from top to bottom). 
     *  This returns a paint you can modify, but that ptr is only valid until 
     *  the next call made to addLayer(). 
     */  
    SkPaint* addLayer(const LayerInfo&);  
  
    /** 
     *  This layer will draw with the original paint, at the specified offset 
     */  
    void addLayer(SkScalar dx, SkScalar dy);  
  
    /** 
     *  This layer will with the original paint and no offset. 
     */  
    void addLayer() { this->addLayer(0, 0); }  
  
    /// Similar to addLayer, but adds a layer to the top.  
    SkPaint* addLayerOnTop(const LayerInfo&);  
  
    /** 
      * Pass list of layers on to newly built looper and return it. This will 
      * also reset the builder, so it can be used to build another looper. 
      */  
    SkLayerDrawLooper* detachLooper();  
  
private:  
    Rec* fRecs;  
    Rec* fTopRec;  
    int  fCount;  
};  

在init()函数中,SkLayerDrawLooper::Builder的对象loopbuilder调用了addLayer()方法。

SkPaint* SkLayerDrawLooper::Builder::addLayer(const LayerInfo& info) {  
    fCount += 1;  
  
    Rec* rec = SkNEW(Rec);  
    rec->fNext = fRecs;  
    rec->fInfo = info;  
    fRecs = rec;  
    if (NULL == fTopRec) {  
        fTopRec = rec;  
    }  
  
    return &rec->fPaint;  
}  

struct Rec {  
        Rec*    fNext;  
        SkPaint fPaint;  
        LayerInfo fInfo;  
    };  

addLayer()函数首先创建一个Rec结构单链表节点,然后把不同的layerInfo插入到该节点中,最后返回该节点中的fPaint。可以看到init()函数中的for循环里会设置这个fPaint的color、style、StrokeWidth和MaskFilter。设置完后loopBuilder使用detachLooper()方法把构造的SkLayerDrawLooper对象交给fLooper成员。

到这里,fLooper中保存了四种不同的paint,因此在onDraw()中调用各种draw api时产生了四种不同图形叠加到一起的效果。

但,在draw api中是draw looper是怎样工作的呢?

可以拿onDraw()中的drawCircle()作为切入点,看一下draw looper的到底是怎样工作的。

void SkCanvas::drawCircle(SkScalar cx, SkScalar cy, SkScalar radius,  
                          const SkPaint& paint) {  
    if (radius < 0) {  
        radius = 0;  
    }  
  
    SkRect  r;  
    r.set(cx - radius, cy - radius, cx + radius, cy + radius);  
    this->drawOval(r, paint);  
}  

void SkCanvas::drawOval(const SkRect& oval, const SkPaint& paint) {  
    SkRect storage;  
    const SkRect* bounds = NULL;  
    if (paint.canComputeFastBounds()) { //判断是否可以快速计算绘制边界(主要判断当前paint和skdrawlooper中的paint是否有mask)  
        bounds = &paint.computeFastBounds(oval, &storage);  
        if (this->quickReject(*bounds)) {  
            return;  
        }  
    }  
  
  
    LOOPER_BEGIN(paint, SkDrawFilter::kOval_Type, bounds)  
  
  
    while (iter.next()) {  
        iter.fDevice->drawOval(iter, oval, looper.paint());  
    }  
  
  
    LOOPER_END  
}  

从上面的代码中可以看出,drawCircle()实际就是drawOval(),通过找出外切矩形来确定圆形的位置和形状。

drawOval()函数可以看出做了三件事情:

  • 计算绘制边界;
  • 外层循环AutoDrawLooper;
  • 内层循环DrawIter。

在第一点中,由于drawOval()的参数中已经有了一个skrect,这可以看做一个初始的绘制边界,之后这个初始边界会被SkDrawLooper中所保存的paint去计算一些变换(比如maskfilter、patheffect),这些变换可能会改变最终的一个绘制边界。如果绘制边界为空,或者为无限,那就拒绝绘制。

第二点,从代码中看LOOPER_BEGIN是一个宏定义,宏展开代码如下:

#define LOOPER_BEGIN(paint, type, bounds)                           \  
    this->predrawNotify();                                          \  
    AutoDrawLooper  looper(this, paint, false, bounds);             \  
    while (looper.next(type)) {                                     \  
        SkAutoBounderCommit ac(fBounder);                           \  
        SkDrawIter          iter(this);  
  
#define LOOPER_END    }  

宏展开后就可以很清楚的看到第一层循环,该循环的判断条件是AutoDrawLooper对象,先看一下这个类的构造函数:

AutoDrawLooper(SkCanvas* canvas, const SkPaint& paint,  
               bool skipLayerForImageFilter = false,  
               const SkRect* bounds = NULL) : fOrigPaint(paint) {  
    fCanvas = canvas;  
    fFilter = canvas->getDrawFilter();  
    fPaint = NULL;  
    fSaveCount = canvas->getSaveCount();  
    fDoClearImageFilter = false;  
    fDone = false;  
  
    if (!skipLayerForImageFilter && fOrigPaint.getImageFilter()) {  
        SkPaint tmp;  
        tmp.setImageFilter(fOrigPaint.getImageFilter());  
        (void)canvas->internalSaveLayer(bounds, &tmp, SkCanvas::kARGB_ClipLayer_SaveFlag,  
                                        true, SkCanvas::kFullLayer_SaveLayerStrategy);  
        // we'll clear the imageFilter for the actual draws in next(), so  
        // it will only be applied during the restore().  
        fDoClearImageFilter = true;  
    }  
  
    if (SkDrawLooper* looper = paint.getLooper()) {  
        void* buffer = fLooperContextAllocator.reserveT<SkDrawLooper::Context>(  
                looper->contextSize());  
        fLooperContext = looper->createContext(canvas, buffer);  
        fIsSimple = false;  
    } else {  
        fLooperContext = NULL;  
        // can we be marked as simple?  
        fIsSimple = !fFilter && !fDoClearImageFilter;  
    }  
}  

在构造函数中可以直接去看第二个if语句,这个语句里所做的事情是:如果paint设置了SkDrawLooper对象,则会在给定的一块buffer创建一个context。如果paint设置的DrawLooper对象是SkLayerDrawLooper对象,则创建的context实际是LayerDrawLooperContext。在构造LayerDrawLooperContext时,它的成员是一个Rec结构指针fCurrRec,fCurrRec会指向paint中的SkLayerDrawLooper对象中的Rec结构链表头。

我们再来看一下SkLayerDrawLooper中Rec这个结构体:

struct Rec {  
    Rec*    fNext;  
    SkPaint fPaint;  
    LayerInfo fInfo;  
};  

Rec链表节点保存着一个layerinfo和一个paint,其中layerinfo结构如下:

/** 
 *  Info for how to apply the layer's paint and offset. 
 * 
 *  fColorMode controls how we compute the final color for the layer: 
 *      The layer's paint's color is treated as the SRC 
 *      The draw's paint's color is treated as the DST 
 *      final-color = Mode(layers-color, draws-color); 
 *  Any SkXfermode::Mode will work. Two common choices are: 
 *      kSrc_Mode: to use the layer's color, ignoring the draw's 
 *      kDst_Mode: to just keep the draw's color, ignoring the layer's 
 */  
struct SK_API LayerInfo {  
    BitFlags            fPaintBits;  
    SkXfermode::Mode    fColorMode;  
    SkVector            fOffset;  
    bool                fPostTranslate; //!< applies to fOffset  

对于layerinfo成员fColorMode的解释是:这个成员用来计算当前layer(这个layer指的的是效果层)的最终颜色,如果这个成员值为kSrc_Mode,则使用当前layer’s paint的颜色,且忽略要绘制layer’s paint的颜色;如果值为kDst_Mode,行为相反。

对于成员fPaintBits,它的有关解释在以下枚举结构中:

/** 
     *  Bits specifies which aspects of the layer's paint should replace the 
     *  corresponding aspects on the draw's paint. 
     *  kEntirePaint_Bits means use the layer's paint completely. 
     *  0 means ignore the layer's paint... except for fColorMode, which is 
     *  always applied. 
     */  
    enum Bits {  
        kStyle_Bit      = 1 << 0,   //!< use this layer's Style/stroke settings  
        kTextSkewX_Bit  = 1 << 1,   //!< use this layer's textskewx  
        kPathEffect_Bit = 1 << 2,   //!< use this layer's patheffect  
        kMaskFilter_Bit = 1 << 3,   //!< use this layer's maskfilter  
        kShader_Bit     = 1 << 4,   //!< use this layer's shader  
        kColorFilter_Bit = 1 << 5,  //!< use this layer's colorfilter  
        kXfermode_Bit   = 1 << 6,   //!< use this layer's xfermode  
  
        /** 
         *  Use the layer's paint entirely, with these exceptions: 
         *  - We never override the draw's paint's text_encoding, since that is 
         *    used to interpret the text/len parameters in draw[Pos]Text. 
         *  - Color is always computed using the LayerInfo's fColorMode. 
         */  
        kEntirePaint_Bits = -1  
  
    };  

fPaintBits用来判断使用当前layer的Style/patheffect/maskfilter/shader/colorfilter/xfermode,还是使用即将要绘制的layer’s paint。

对于成员fOffset和fPostTranslate,它们用来处理当前layer的位置偏移,会改变canvas的matrix。

因此,layerinfo保存了SkLayerDrawLooper中的每一个layer的paint mode flag和偏移信息。

然后回到之前的外层循环宏展开,构造完AutoDrawLooper对象looper,就会执行looper.next(type)。

bool next(SkDrawFilter::Type drawType) {  
    if (fDone) {  
        return false;  
    } else if (fIsSimple) {  
        fDone = true;  
        fPaint = &fOrigPaint;  
        return !fPaint->nothingToDraw();  
    } else {  
        return this->doNext(drawType);  
    }  
}  

bool AutoDrawLooper::doNext(SkDrawFilter::Type drawType) {  
    fPaint = NULL;  
    SkASSERT(!fIsSimple);  
    SkASSERT(fLooperContext || fFilter || fDoClearImageFilter);  
  
    SkPaint* paint = fLazyPaint.set(fOrigPaint);  
  
    if (fDoClearImageFilter) {  
        paint->setImageFilter(NULL);  
    }  
  
    if (fLooperContext && !fLooperContext->next(fCanvas, paint)) {  
        fDone = true;  
        return false;  
    }  
    if (fFilter) {  
        if (!fFilter->filter(paint, drawType)) {  
            fDone = true;  
            return false;  
        }  
        if (NULL == fLooperContext) {  
            // no looper means we only draw once  
            fDone = true;  
        }  
    }  
    fPaint = paint;  
  
    // if we only came in here for the imagefilter, mark us as done  
    if (!fLooperContext && !fFilter) {  
        fDone = true;  
    }  
  
    // call this after any possible paint modifiers  
    if (fPaint->nothingToDraw()) {  
        fPaint = NULL;  
        return false;  
    }  
    return true;  
}  

考虑looper.next(type)执行到AutoDrawLooper::doNext()的情况,在doNext()第二个if语句中,会去执行fLooperContext->next(fCanvas, paint),这里执行的就是刚刚构造的LayerDrawLooperContext对象中的next()方法:

bool SkLayerDrawLooper::LayerDrawLooperContext::next(SkCanvas* canvas,  
                                                     SkPaint* paint) {  
    canvas->restore();  
    if (NULL == fCurrRec) {  
        return false;  
    }  
  
    ApplyInfo(paint, fCurrRec->fPaint, fCurrRec->fInfo);  
  
    canvas->save();  
    if (fCurrRec->fInfo.fPostTranslate) {  
        postTranslate(canvas, fCurrRec->fInfo.fOffset.fX,  
                      fCurrRec->fInfo.fOffset.fY);  
    } else {  
        canvas->translate(fCurrRec->fInfo.fOffset.fX,  
                          fCurrRec->fInfo.fOffset.fY);  
    }  
    fCurrRec = fCurrRec->fNext;  
  
    return true;  
}  

看到这里就比较明显了,我们在drawLooper.cpp中的init()函数中定义的四种效果会在这里进行处理。首先是在ApplyInfo()中处理我们定义的color、style、width和maskfilter;然后处理offset;最后fCurrRec指向下一个Rec节点。如果到了Rec链表尾,则外层循环结束。看一下ApplyInfo()设置的info的过程:

void SkLayerDrawLooper::LayerDrawLooperContext::ApplyInfo(  
        SkPaint* dst, const SkPaint& src, const LayerInfo& info) {  
  
    dst->setColor(xferColor(src.getColor(), dst->getColor(), info.fColorMode));  
  
    BitFlags bits = info.fPaintBits;  
    SkPaint::TextEncoding encoding = dst->getTextEncoding();  
  
    if (0 == bits) {  
        return;  
    }  
    if (kEntirePaint_Bits == bits) {  
        // we've already computed these, so save it from the assignment  
        uint32_t f = dst->getFlags();  
        SkColor c = dst->getColor();  
        *dst = src;  
        dst->setFlags(f);  
        dst->setColor(c);  
        dst->setTextEncoding(encoding);  
        return;  
    }  
  
    if (bits & kStyle_Bit) {  
        dst->setStyle(src.getStyle());  
        dst->setStrokeWidth(src.getStrokeWidth());  
        dst->setStrokeMiter(src.getStrokeMiter());  
        dst->setStrokeCap(src.getStrokeCap());  
        dst->setStrokeJoin(src.getStrokeJoin());  
    }  
  
    if (bits & kTextSkewX_Bit) {  
        dst->setTextSkewX(src.getTextSkewX());  
    }  
  
    if (bits & kPathEffect_Bit) {  
        dst->setPathEffect(src.getPathEffect());  
    }  
    if (bits & kMaskFilter_Bit) {  
        dst->setMaskFilter(src.getMaskFilter());  
    }  
    if (bits & kShader_Bit) {  
        dst->setShader(src.getShader());  
    }  
    if (bits & kColorFilter_Bit) {  
        dst->setColorFilter(src.getColorFilter());  
    }  
    if (bits & kXfermode_Bit) {  
        dst->setXfermode(src.getXfermode());  
    }  
  
    // we don't override these  
#if 0  
    dst->setTypeface(src.getTypeface());  
    dst->setTextSize(src.getTextSize());  
    dst->setTextScaleX(src.getTextScaleX());  
    dst->setRasterizer(src.getRasterizer());  
    dst->setLooper(src.getLooper());  
    dst->setTextEncoding(src.getTextEncoding());  
    dst->setHinting(src.getHinting());  
#endif  
}  

对于这行:dst->setColor(xferColor(src.getColor(), dst->getColor(), info.fColorMode));

dst指的是draw api中的paint(跟着函数调用一层层传下来),也就是即将要绘制的layer’s paint ;src指的是SkLayerDrawLooper中Rec结构成员中的paint。dst设置当前paint的颜色时是根据layerinfo成员fColorMode决定的(如上面layerinfo中的注释)。

我们回到DrawLooper这个例子,只拿绘制的圆形来说明:gParams数组定义的每组效果的颜色依次是白色,红色,蓝色,灰色;绘制圆形时先绘制白色的圆环(style=stroke),然后时红色的圆环(style=stroke),之后是蓝色的圆盘(style=full),最后是灰色的圆盘(style=full),这里每次绘制都是绘制到一个layer上;由于每组效果的layerinfo成员fColorMode都设置的是kSrc_mode,因此这些layer上的图案混合在一起的时候,在相互重叠的地方都保持的是绘制时当前layer的颜色。直观的效果看上去就是后面绘制的图案被之前的layer的图案挡住,白色圆环盖在了红色圆环上,蓝色圆盘的边缘被上面两层图案盖住,灰色圆盘被之前三层的图案盖住。

下面我们再看内存循环,先看SkDrawIter的构造函数:

SkDrawIter(SkCanvas* canvas, bool skipEmptyClips = true) {  
    canvas = canvas->canvasForDrawIter();  
    fCanvas = canvas;  
    canvas->updateDeviceCMCache();  
  
    fClipStack = &canvas->fClipStack;  
    fBounder = canvas->getBounder();  
    fCurrLayer = canvas->fMCRec->fTopLayer;  
    fSkipEmptyClips = skipEmptyClips;  
}  

对于SkDrawIter类,它的基类是SkDraw;它的会在构造函数中为每一层layer(这个layer指的是图层)更新相对应的MCRec状态(图层链表DeviceCM中每一个layer与状态栈中的栈帧MCRec有着一一对应关系,但有的栈帧MCRec可能没有layer);这是为了在正式绘制在layer上之前,调整好layer的空间关系(matrix)和剪裁区域(clip),后面正式开始绘制的时候都按照调整好的matrix和clip去绘制。

内存循环的判断条件是iter.next():

bool next() {  
    // skip over recs with empty clips  
    if (fSkipEmptyClips) {  
        while (fCurrLayer && fCurrLayer->fClip.isEmpty()) {  
            fCurrLayer = fCurrLayer->fNext;  
        }  
    }  
  
    const DeviceCM* rec = fCurrLayer;  
    if (rec && rec->fDevice) {  
  
        fMatrix = rec->fMatrix;  
        fClip   = &((SkRasterClip*)&rec->fClip)->forceGetBW();  
        fRC     = &rec->fClip;  
        fDevice = rec->fDevice;  
        fBitmap = &fDevice->accessBitmap(true);  
        fPaint  = rec->fPaint;  
        SkDEBUGCODE(this->validate();)  
  
        fCurrLayer = rec->fNext;  
        if (fBounder) {  
            fBounder->setClip(fClip);  
        }  
        // fCurrLayer may be NULL now  
  
        return true;  
    }  
    return false;  
}  

SkDrawIter类的next()方法的作用是:在正式绘制每一层layer之前,首先跳过clip为空的layer(即clip为空的layer不绘制);然后把当前要绘制的layer一些有用参数传递给SkDrawIter对象的成员,这些成员都是已经更新过matrix和clip状态的,已经具备了绘制条件;最后判断是否到了图层链表尾,用于内层循环判断条件。

从代码中看出内层循环依然是对layer的MC状态一些迭代更新,并在循环体中调用实际绘制函数去绘制当前状态所依附的所有Layer,这里与SkDrawLooper没有关系。

总结:

看到这里可以对SkDrawLooper(针对子类SkLayerDrawLooper)的作用作以下总结:

  1. paint可比喻为画笔,画笔可以画出各种效果;这些效果会分布在不同的效果层。SkLayerDrawLooper::LayerInfo定义效果层的paint mode flag和偏移;它决定了在绘制前使用当前效果层的paint效果还是使用即将绘制的paint效果,每一个paint对应的它对应的效果用Rec节点保存在SkLayerDrawLooper中,这个Rec结构可以认为是一个效果层。
  2. SkLayerDrawLooper::Builder的对象调用addLayer()函数首先创建Rec结构单链表节点,然后把不同的layerInfo插入到该节点中,最后返回每个节点中与新添加的layerinfo对应的fPaint。有了Rec结构链表,SkLayerDrawLooper::Builder会调用detachLooper()方法返回一个SkLayerDrawLooper对象,这个SkLayerDrawLooper对象可以设置到即将绘制的paint中。这里的addLayer就是添加效果层。
  3. 把SkLayerDrawLooper对象设置给一个paint,当canvas调用draw api时会使用SkLayerDrawLooper对象去计算绘制边界,然后在draw api的外层循环中使用SkLayerDrawLooper::LayerDrawLooperContext::next()函数去判断使用即将绘制的paint效果还是looper中paint效果,并且会处理每一层的偏移。

对于二层循环总结如下:

  1. 外层循环的作用是判断使用即将绘制的paint效果还是looper中paint效果,并且会处理每一层的偏移;
  2. 内层循环是在正式绘制在layer(这个layer是图层)上之前,调整好layer的空间关系(matrix)和剪裁区域(clip),然后跳过clip为空的layer,把当前要绘制的layer一些有用参数传递给SkDrawIter对象的成员,后面让SkDrawIter中的Device去调用实际的绘制函数;这个过程依次迭代。

假设SkLayerDrawLooper对象为looper,SkDrawIter对象为iter,下面这张图简单的描述了两层循环的行为,绿色虚线内为一次外层循环,红色虚线为一次内层循环。

looper