SkRect is basic to many drawing and measuring operations. It can be drawn using canvas.drawRect(), but it is also used to return the bounds of objects like paths and text characters. It is specified using SkScalar values.

SkIRect is the integer counter part to SkRect, but is specified using 32bit integers.

struct SkRect {
   SkScalar fLeft;
   SkScalar fTop;
   SkScalar fRight;
   SkScalar fBottom;
   // methods

SkRect rect = SkRect::MakeLTRB(left, top, right, bottom);

SkRect has the usual getters, to return width(), height(), centerX(), etc. It also has methods to compute unions and intersections between rectangles.

Converting between SkRect and SkIRect is asymetric. Short of overflow issues when SkScalar is an int, converting from SkIRect to SkRect is straight forward:

SkRect::set(const SkIRect&);

However, convert from SkRect to SkIRect needs to know how to go from fractional values to integers.

SkRect::round(SkIRect*) const;     // Round each coordinate.
SkRect::roundOut(SkIRect*) const;  // Apply floor to left/top,
                                   // and ceil to right/bottom.

In Skia, rectangle coordinates describe the boundary of what is drawn, such that an empty rectangle encloses zero pixels:

bool SkRect::isEmpty() const { return fLeft >= fRight || fTop >= fBottom; }

SkScalar SkRect::width() const { return fRight - fLeft; }

SkScalar SkRect::height() const { return fBottom - fTop; }

bool SkRect::contains(SkScalar x, SkScalar y) const {
    return fLeft <= x && x < fRight && fTop <= y && y < fBottom;

Thus, to draw a single pixel (assuming no matrix on the canvas), the rectangle should be initialized as follows:

SkRect r = SkRect::MakeXYWH(x, y, SkIntToScalar(1), SkIntToScalar(1));

The same conventions hold for the integer counterpart: SkIRect. This also dovetails with SkRegion, which has the same model for set membership, and which uses SkIRect.