How to use multiple viewports in OpenGL

Cross-browser @media (width) and @media (height) values 

const vw = Math.max(document.documentElement.clientWidth || 0, window.innerWidth || 0)
const vh = Math.max(document.documentElement.clientHeight || 0, window.innerHeight || 0)

window.innerWidth and window.innerHeight

• gets CSS viewport @media (width) and @media (height) which include scrollbars
• initial-scale and zoom variations may cause mobile values to wrongly scale down to what PPK calls the visual viewport and be smaller than the @media values
• zoom may cause values to be 1px off due to native rounding
• undefined in IE8-

document.documentElement.clientWidth and .clientHeight

• equals CSS viewport width minus scrollbar width
• matches @media (width) and @media (height) when there is no scrollbar
• same as jQuery(window).width() which jQuery calls the browser viewport
• available cross-browser
• inaccurate if doctype is missing

Resources

• Live outputs for various dimensions
• verge uses cross-browser viewport techniques
• actual uses matchMedia to obtain precise dimensions in any unit

When using an onthogonal projection, you can simply think of the ortho matrix as defining some axis-aligned 2D rectangle in the xy-plane, which describes the area of the scene which is mapped to the viewport. If the aspect ratio of that rectangle does not match the aspect ratio of the viewport, the image will be distored accordingly.

Let us use the following definitions:

V: aspect of the viewport:
  V = viewport_width / viewport_height
P: aspect of the ortho projection:
  P = (right - left) / (top - bottom)
O: aspect of some axis-aligned rectangle which is drawn
  O = (x_max - x_min) / (y_max - y_min)

When the transformations are applied, the object will appear with aspect ratio O / P * V on the screen.

Usually, when speaking of "keeping aspect ratio", we set P == V so that V / P cancels each other out in the above formula, and objects appear with exactly that aspect ratio they are drawn in eye space.

And this is already exactly what you get with your code.:

Then I create a quad going from -1,-1 to 1, 1 so that it's origin is at the center.

That is a square, and I wonder how you ever can expect this to come out as a rectangle when the aspect ratio is preserved.

From your images, it is clear that your object of interest is a rectangle with aspect ratio 2/3. So you should also draw it as a rectangle with such an aspect ratio. There are multiple ways to achieve this. Since the viewport size seems to be a given, you can sill tweak O or P or both.

However, to me it seems like you are completely overcomplicating things. If I got you right, you have some "design space" of 320x480 pixels, which is your "region of interest" which you always want to be visible on the screen, no matter what the viewport size is. To achieve that, you could do the following:

float target_width = 320.f;
float target_height = 480.f;
float A = target_width / target_height; // target aspect ratio 
// ... calculate V as above
if (V >= A) {
    // wide viewport, use full height
    ortho(-V/A * target_width/2.0f, V/A * target_width/2.0f, -target_height/2.0f, target_height/2.0f, ...);
} else {
    // tall viewport, use full width
    ortho(-target_width/2.0f, target_width/2.0f, -A/V*target_height/2.0f, A/V*target_height/2.0f, ...);
}

Now, you can work in your "design range" in pixels. In that example, the always visible range will be from (-160, -240) to (160, 240)`, and if you draw an rectangle exactly with that coordinates, it will match the blue box in your image, in any of the viewports. If you don't want the origin in the center, you can of course translate it.