Pixel perfect: choosing the right correction (OpenGL)

Question

I was trying to draw pixel perfect bitmaps using quads in OpenGL... and to my surprise, a very important row of pixels was missing: (green is ok, red is bad)

The sizes of the quads were 30x30 px, with the red 2x2 in the top left corner.

• Can someone explain what is happening?
• How does OpenGL decide where to put pixels?
• Which correction should I choose?

The texture uses GL_NEAREST.

    glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);

I don't use glViewport and glOrtho. Instead, I send the screen dimensions through uniforms to the shaders:

uniform float cw;
uniform float ch;
...
gl_Position = vec4(     -1.0f + 2.0f*(in_pos.x + in_offset.x)/cw, 
                         1.0f - 2.0f*(in_pos.y + in_offset.y)/ch,
                         0.0f, 1.0f);

But that doesn't seem to be the source of the problem because when I add (0.5,0.5) in the shader (equivalent to the 3rd case), the problem goes away...

Ok, now I realize that the third option doesn't work either:

Answer 1

Check that you've passed correct screen size into shader.

When you create window, you define its outer size, but the client area, where the rendering context is put, can be smaller when window has header/decorations.

So you should either ask the client area size with GetClientRect(), or find the "correct" window size by calling AdjustWindowRect() before creating window - this would give you the window size that will have client area of the size you wanted initially.