How do glPushMatrix() and glPopMatrix() keep the scene the same?

Question

I found some code online which will move a box across the screen, then reset it after the box hits the end of the screen.
Here is the code:

Answer 1

It's only the matrix, used to transform coordinates, that is restored by glPopMatrix. Not the whole framebuffer. The framebuffer contains the rendering of the quadrilateral, changing the matrix afterward doesn't affect anything already rendered.

Answer 2

OpenGL is a drawing API. When you call drawing functions, things are literally drawn to the framebuffer the very moment you issue the draw call -- well actually OpenGL batches up all the commands internally and processes them in order. But when OpenGL is about to process those drawing calls, it will draw to the framebuffer with the matrices set to the state at this specific position in the batch.

So to reemphase this: OpenGL does not do any kind of scene management, it just draws things to the framebuffer in the order and way you issue the drawing commands. You send a triangle: OpenGL will transform and draw it. There's no scene internally built. Once you understood this it becomes trivial to understand how the matrix stack can do its "magic".

Answer 3

glPushMatrix duplicates the matrix on top of the stack (you're always working with the top one). Any other transformation you are doing modifies this top matrix, the duplicated one. When you do glPopMatrix we are back to the original matrix.

For example, suppose you want to draw a car. You set up the matrix to draw the body of the car, let's call it M1. Now, you want to draw one wheel. You can either compute M2 - the matrix needed for the wheel to be displayed correctly - or, since the wheel is relative to the body of the car (thus, there is a matrix M3 such that M2 = M1 * M3) you modify M1. But the car has 4 wheels, you need to keep a copy of M1. You do this by doing a glPushMatrix, you get back the copy by doing a glPopMatrix.

When you draw anything on the screen, you are giving coordinates in object space. To really display something, those coordinates need to be transformed. For that we have some matrices.

In the wheel example, you have only one wheel geometry but because you are using different matrices, there will be four wheels drawn. glPushMatrix and glPopMatrix work only with the matrix, the actual vertex data is kept into the GPU, each glVertex sends another one there and it cannot be removed. See the following image, the matrices are used only for transforming object coordinates to world coordinates (actually, all matrices can be pushed into a stack)